WO2003071743A1

WO2003071743A1 - Software, systems and methods for managing a distributed network

Info

Publication number: WO2003071743A1
Application number: PCT/US2003/004648
Authority: WO
Original assignee: Centrisoft Corporation
Priority date: 2002-02-18
Filing date: 2003-02-18
Publication date: 2003-08-28
Also published as: AU2003217532A1; TW200304298A

Abstract

A system and method for managing network bandwidth consumption. The system may include an agent module loadable on a networked computer and configured to aid in managing bandwidth consumption within a network. The agent module is configured to obtain an allocation of network bandwidth usable by the networked computer, and is further configured to sub-allocate such allocation among multiple bandwidth-consuming components associated with the networked computer. The system may further include multiple such agent modules loadable on plural networked computers, and a control module configured to interact with each of the agent modules to dynamically manage bandwidth usage by the networked computers.

Description

SOFTWARE, SYSTEMS AND METHODS FOR MANAGING A DISTRIBUTED NETWORK

Cross-Reference to Related Applications

The application is based upon and claims the benefit of U.S. provisional patent application Serial No. 60/357,731, filed February 18, 2002.

Technical Field

The present invention relates generally to distributed network systems, and more particularly to software, systems and methods for managing resources of a distributed network. Background

Both public and private networks have shifted toward a predominantly distributed computing model, and have grown steadily in size, power and complexity. This growth has been accompanied by a corresponding increase in demands placed on information technology to increase enterprise-level productivity, operations and customer/user support. To achieve interoperability in increasingly complex network systems, TCP/IP and other standardized communication protocols have been aggressively deployed. Although many of these protocols have been effective at achieving interoperability, their widespread deployment has not been accompanied by a correspondingly aggressive development of management solutions for networks using these protocols. Indeed, conventional computer networks provide little in the way of solutions for managing network resources, and instead typically provide what is known as "best efforts" service to all network traffic. Best efforts is the default behavior of TCP/IP networks, in which network nodes simply drop packets indiscriminately when faced with excessive network congestion. With best efforts service, no mechanism is provided to avoid the congestion that leads to dropped packets, and network traffic is not categorized to ensure reliable delivery of more important data. Also, users are not provided with information about network conditions or underperforming resources. This lack of management frequently results in repeated, unsuccessful network requests, user frustration and diminished productivity. Problems associated with managing network resources are intensified by the dramatic increase in the demand for these resources. New applications for use in distributed networking environments are being developed at a rapid pace. These applications have widely varying performance requirements. Multimedia applications, for example, have a very high sensitivity to jitter, loss and delay. By contrast, other types of applications can tolerate significant lapses in network performance. Many applications, particularly continuous media applications, have very high bandwidth requirements, while others have bandwidth requirements that are comparatively modest. A further problem is that many bandwidth- intensive applications are used for recreation or other low priority tasks. hi the absence of effective management tools, the result of this increased and varied competition for network resources is congestion, application unpredictability, user frustration and loss of productivity. When networks are unable to distinguish unimportant tasks or requests from those that are mission critical, network resources are often used in ways that are inconsistent with business objectives. Bandwidth may be wasted or consumed by low priority tasks. Customers may experience unsatisfactory network performance as a result of internal users placing a high load on the network.

Various solutions have been employed, with limited success, to address these network management problems. For example, to alleviate congestion, network managers often add more bandwidth to congested links. This solution is expensive and can be temporary - network usage tends to shift and grow such that the provisioned link soon becomes congested again. This often happens where the underlying cause of the congestion is not addressed. Usually, it is desirable to intelligently manage existing resources, as opposed to "over- provisioning," i.e. simply providing more resources to reduce scarcity.

A broad, conceptual class of management solutions may be thought of as attempts to increase "awareness" in a distributed networking environment. The concept is that where the network is more aware of applications or other tasks running on networked devices, and vice versa, then steps can be taken to make more efficient use of network resources. For example, if network management software becomes aware that a particular user is running a low priority application, then the software could block or limit that user's access to network resources. If management software becomes aware that the network population at a given instance includes a high percentage of outside customers, bandwidth preferences and priorities could be modified to ensure that the customers had a positive experience with the network. In the abstract, increasing application and network awareness is a desirable goal, however application vendors largely ignore these considerations and tend to focus not on network infrastructure, but rather on enhancing application functionality. Quality of service ("QoS") and policy-based management techniques represent efforts to bridge the gap between networks, applications and users in order to more efficiently manage the use of network resources. QoS is a term referring to techniques which allow network-aware applications to request and receive a predictable level of service in terms of performance specifications such as bandwidth, jitter, delay and loss. Known QoS methods include disallowing certain types of packets, slowing transmission rates, establishing distinct classes of services for certain types of packets, marking packets with a priority value, and various queuing methods. In a distributed environment having scarce resources, QoS techniques necessarily introduce unfairness into the system by giving preferential treatment to certain network traffic. Policy-based network management uses policies, or rules, to define how network resources are to be used. In a broad sense, a policy includes a condition and an action. An example of a policy could be to block access or disallow packets (action) if the IP source address of the data is included on a list of disallowed addresses (condition). One use of policy-based network management techniques is to determine when and how the unfairness introduced by QoS methods should apply.

Policy-based management solutions typically require that network traffic be classified before it is acted upon. The classification process can occur at various levels of data abstraction, and may be described in terms of layered communication protocols that network devices use to communicate across a network link. There are two protocol layering models which dominate the field. The first is the OSI reference model, depicted in Fig. 1. The layers of the OSI model are: application (layer 7), presentation (layer 6), session (layer 5), transport (layer 4), network (layer 3), data link (layer 2) and physical (layer 1). The second major model forms the basis for the TCP/IP protocol suite. Its layers are application, transport, network, data link and hardware, as also depicted in Fig. 1. The TCP/IP layers correspond in function to the OSI layers, but without a presentation or session layer. In both models, data is processed and changes form as it is sequentially passed between the layers.

Known policy based management solutions and QoS methods typically classify data by monitoring data flows at the transport layer and below. For example, a common multi- parameter classifier is the well known "five-tuple" consisting of (IP source address, IP destination address, IP protocol, TCP/UDP source port and TCP/UDP destination port). These parameters are all obtained at the transport and network layers of the models. The large majority of existing policy-based, QoS solutions are implemented by monitoring and classifying network activity at these protocol layers. However, the higher the protocol layer, the more definitive and specific the available data and classifiers. Because conventional policy-based, QoS systems do not employ classifiers at higher than the transport layer, they cannot employ policy-based techniques or QoS methods using the richer and more detailed data available at the higher layers. The conventional systems are thus limited in their ability to make the network more application-aware and vice versa.

In addition, the known systems for managing network resources do not effectively address the problem of bandwidth management. Bandwidth is often consumed by low priority tasks at the expense of business critical applications. In systems that do provide for priority based bandwidth allocations, the bandwidth allocations are static and are not adjusted dynamically in response to changing network conditions.

Summary Accordingly, the present description provides systems and methods for managing network bandwidth consumption, which may include use of an agent module loadable on a networked computer. The agent module is configured to obtain an allocation of network bandwidth usable by the networked computer, and is further configured to sub-allocate such allocation among multiple bandwidth-consuming components associated with the networked computer. The system may further include multiple such agent modules loaded on plural networked computers, and a control module configured to interact with each of the agent modules to dynamically manage bandwidth usage by the networked computers.

Brief Description of the Drawings Fig. 1 is a conceptual depiction of the OSI and TCP/IP layered protocol models. Fig. 2 is a view of a distributed network system in which the software, systems and methods described herein may be deployed.

Fig. 3 is a schematic view of a computing device that may be deployed in the distributed network system of Fig. 2.

Fig. 4 is a block diagram view depicting exemplary agent modules and control modules that may be used to manage resources in a distributed network such as that depicted in Fig. 2.

Fig. 5 is a block diagram view depicting various exemplary components that may employed in connection with the described software, systems and methods, including two agent modules, a control module and a configuration utility. Fig. 6 is a block diagram depicting an exemplary deployment of an agent module in relation to a layered protocol stack of a computing device.

Fig. 7 is a block diagram depicting another exemplary deployment of an agent module in relation to a layered protocol stack of a computing device.

Fig. 8 is a block diagram depicting yet another exemplary deployment of an agent module in relation to a layered protocol stack of a computing device.

Fig. 9 is a block diagram depicting exemplary component parts of an agent module embodiment.

Fig. 10 is a block diagram depicting exemplary component parts of a control module embodiment. Fig. 11A is a flowchart depicting a method for allocating bandwidth among a plurality of computers.

Fig. 1 IB is a flowchart depicting another method for allocating bandwidth among a plurality of computers. Fig. 11C is a flowchart depicting yet another method for allocating bandwidth among a plurality of computers.

Fig. 1 ID is a flowchart depicting yet another method for allocating bandwidth among a plurality of computers.

Fig. 12 is a flowchart depicting a method for monitoring the status of network resources.

Fig. 13 is a view of a main configuration screen of a configuration utility that may be employed in connection with the software, systems and methods described herein.

Fig. 14 is a view of another configuration screen of the configuration utility depicted in Fig. 13. Fig. 15 is a view of yet another configuration screen of the configuration utility depicted in Fig. 13.

Fig. 16 is a view of yet another configuration screen of the configuration utility depicted in Fig. 13.

Fig. 17 schematically depicts a multi-tier control architecture that may be employed in connection with the management software, systems and methods described herein.

Fig. 18 depicts an exemplary network configuration allowing for centralized definition of network policies, and dissemination of those policies to pertinent distributed locations.

Fig. 19 depicts an exemplary multi-tier system and method for providing granular control over bandwidth usage of individual bandwidth-consuming components.

Fig. 20 depicts an exemplary computing device loaded with an embodiment of an agent software module configured to manage bandwidth consumption of an individual application socket or transaction.

Fig. 21 depicts an exemplary computing device loaded with alternate embodiments of an agent module configured to monitor and manage network traffic to and from applications running on the computing device.

Detailed Description The present description provides a system and method for managing network resources in a distributed networldng environment, such as distributed network 10 depicted in Fig. 2. The software, system and methods increase productivity and customer/user satisfaction, minimize frustration associated with using the network, and ultimately ensure that network resources are used in a way consistent with underlying business or other objectives.

The systems and methods may employ two main software components, an agent and a control module, also referred to as a control point. The agents and control points may be deployed throughout distributed network 10, and may interact with each other to manage network resources. A plurality of agents may be deployed to intelligently couple clients, servers and other computing devices to the underlying network. The deployed agents monitor, analyze and act upon network events relating to the networked devices with which they are associated. The agents typically are centrally coordinated and/or controlled by one or more control points. The agents and control points may interact to control and monitor network events, track operational and congestion status of network resources, select optimum targets for network requests, dynamically manage bandwidth usage, and share information about network conditions with customers, users and IT personnel. As indicated, distributed network 10 may include a local network 12 and a plurality of remote networks 14 linked by a public network 16 such as the Internet. The local network and remote networks may be connected to the public network with network infrastructure devices such as routers 18.

Local network 12 typically includes servers 20 and client devices such as client computers 22 interconnected by network link 24. Additionally, local network 12 may include any number and variety of devices, including file servers, applications servers, mail servers, WWW servers, databases, client computers, remote access devices, storage devices, printers and network infrastructure devices such as routers, bridges, gateways, switches, hubs and repeaters. Remote networks 14 may similarly include any number and variety of networked devices.

Indeed, virtually any type of computing device may be connected to the networks depicted in Fig. 2, including general purpose computers, laptop computers, handheld computers, wireless computing devices, mobile telephones, pagers, pervasive computing devices and various other specialty devices. Typically, many of the connected devices are general purpose computers which have at least some of the elements shown in Fig. 3, a block diagram depiction of a computer system 40. Computer system 40 includes a processor 42 that processes digital data. The processor may be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor implementing a combination of instruction sets, a microcontroller, or virtually any other processor/controller device. The processor may be a single device or a plurality of devices.

Referring still to Fig. 3, it will be noted that processor 42 is coupled to a bus 44 which transmits signals between the processor and other components in the computer system. Those skilled in the art will appreciate that the bus may be a single bus or a plurality of buses. A memory 46 is coupled to bus 44 and comprises a random access memory (RAM) device 47 (referred to as main memory) that stores information or other intermediate data during execution by processor 42. Memory 46 also includes a read only memory (ROM) and/or other static storage device 48 coupled to the bus that stores information and instructions for processor 42. A basic input/output system (BIOS) 49, containing the basic routines that help to transfer information between elements of the computer system, such as during start-up, is stored in ROM 48. A data storage device 50 also is coupled to bus 44 and stores information and instructions. The data storage device may be a hard disk drive, a floppy disk drive, a CD- ROM device, a flash memory device or any other mass storage device. In the depicted computer system, a network interface 52 also is coupled to bus 44. The network interface operates to connect the computer system to a network (not shown).

Computer system 40 may also include a display device controller 54 coupled to bus 44. The display device controller allows coupling of a display device to the computer system and operates to interface the display device to the computer system. The display device controller 54 may be, for example, a monochrome display adapter (MDA) card, a color graphics adapter (CGA) card, or other display device controller. The display device (not shown) may be a television set, a computer monitor, a flat panel display or other display device. The display device receives information and data from processor 42 through display device controller 54 and displays the information and data to the user of computer system 40. An input device 56, including alphanumeric and other keys, typically is coupled to bus 44 for communicating information and command selections to processor 42. Alternatively, input device 56 is not directly coupled to bus 44, but interfaces with the computer system via infra-red coded signals transmitted from the input device to an infra-red receiver in the computer system (not shown). The input device may also be a remote control unit having keys that select characters or command selections on the display device.

The various computing devices coupled to the networks of Fig. 2 typically communicate with each other across network links using communications software employing various communications protocols. The communications software for each networked device typically consists of a number of protocol layers, through which data is sequentially transferred as it is exchanged between devices across a network link. Fig. 1 respectively depicts the OSI layered protocol model and a layered model based on the TCP/IP suite of protocols. These two models dominate the field of network communications software. As seen in the figure, the OSI model has seven layers, including an application layer, a presentation layer, a session layer, a transport layer, a network layer, a data link layer and a physical layer. The TCP/IP -based model includes an application layer, a transport layer, a network layer, a data link layer and a physical layer.

Each layer in the models plays a different role in network communications. Conceptually, all of the protocol layers lie in a data transmission path that is "between" an application program running on the particular networked device and the network link, with the application layer being closest to the application program. When data is transferred from an application program running on one computer across the network to an application program running on another computer, the data is transferred down through the protocol layers of the first computer, across the network link, and then up through the protocol layers on the second computer. In both of the depicted models, the application layer is responsible for interacting with an operating system of the networked device and for providing a window for application programs running on the device to access the network. The transport layer is responsible for providing reliable, end-to-end data transmission between two end points on a network, such as between a client device and a server computer, or between a web server and a DNS server. Depending on the particular transport protocol, transport functionality may be realized using either connection-oriented or connectionless data transfer. The network layer typically is not concerned with end-to-end delivery, but rather with forwarding and routing data to and from nodes between endpoints. The layers below the transport and network layers perform other functions, with the lowest levels addressing the physical and electrical issues of transmitting raw bits across a network link.

The systems and methods described herein are applicable to a wide variety of network environments employing communications protocols adhering to either of the layered models depicted in Fig. 1, or to any other layered model. Furthermore, the systems and methods are applicable to any type of network topology, and to networks using both physical and wireless connections.

The present description provides software, systems and methods for managing the resources of an enterprise network, such as that depicted in Fig. 2. This may be accomplished using two interacting software components, an agent and a control point, both of which may be adapted to run on, or be associated with, computing devices such as the computing device described with reference to Fig. 3. As seen in Fig. 4, a plurality of agents 70 and one or more control points 72 may be deployed throughout distributed network 74 by loading the agent and control point software modules on networked computing devices such as clients 22 and server 20. As will be discussed in detail, the agents and control points may be adapted and configured to enforce system policies; to monitor and analyze network events, and take appropriate action based on these events; to provide valuable information to users of the network; and ultimately to ensure that network resources are efficiently used in a manner consistent with underlying business or other goals.

The described software, systems and methods may be configured using a third software component, to be later discussed in more detail with reference to Figs. 5 and 13 - 16. Typically, this configuration utility is a platform-independent application that provides a graphical user interface for centrally managing configuration information for the control points and agents. In addition, the configuration utility may be adapted to communicate and interface with other management systems, including management platforms supplied by other vendors. As indicated in Fig. 4, each control point 72 is typically associated with multiple agents 70, and the associated agents are referred to as being within a domain 76 of the particular control point. The control points coordinate and control the activity of the distributed agents within their domains, hi addition, the control points may monitor the status of network resources, and share this information with management and support systems and with the agents.

Control points 72 and agents 70 may be flexibly deployed in a variety of configurations. For example, each agent may be associated with a primary control point and one or more backup control points that will assume primary control if necessary. Such a configuration is illustrated in Fig. 4, where control points 72 within the dashed lines function as primary connections, with the control point associated with server device 20 functioning as a backup connection for all of the depicted agents. In addition, the described exemplary systems may be configured so that one control point coordinates and controls the activity of a single domain, or of multiple domains. Alternatively, one domain may be controlled and coordinated by the cooperative activity of multiple control points. In addition, agents may be configured to have embedded control point functionality, and may therefore operate without an associated control point entity.

Typically, the agents monitor network resources and the activity of the device with which they are associated, and communicate this information to the control points, hi response to monitored network conditions and data reported by agents, the control points may alter the behavior of particular agents in order to provide the desired network services. The control points and agents may be loaded on a wide variety of devices, including general purpose computers, servers, routers, hubs, palm computers, pagers, cellular telephones, and virtually any other networked device having a processor and memory. Agents and control points may reside on separate devices, or simultaneously on the same device. Fig. 5 illustrates an example of the way in which the various components of the described software, systems and methods may be physically interconnected with a network link 90. The components are all connected to network link 90 by means of layered communications protocol software 92. The components communicate with each other via the communications software and network link. As will be appreciated by those skilled in the art, network link 90 may be a physical or wireless connection, or a series of links including physical and wireless segments. More specifically, the depicted system includes an agent 70 associated with a client computing device 22, including an application program 98. Another agent is associated with server computing device 20. The agents monitor the activity of their associated computing devices and communicate with control point 72. Configuration utility 106 communicates with all of the other components, and with other management systems, to configure the operation of the various components and monitor the status of the network.

The system policies that define how network resources are to be used may be centrally defined and tailored to most efficiently achieve underlying goals. Defined policies are accessed by the control points, which in turn communicate various elements and parameters associated with the policies to the agents within their domain. At a very basic level, a policy contains rules about how network resources are to be used, with the rules containing conditions and actions to be taken when the conditions are satisfied. The agents and control points monitor the network and devices connected to the network to determine when various rules apply and whether the conditions accompanying those rules are satisfied. Once the agents and/or control points determine that action is required, they take the necessary action(s) to enforce the system policies.

For example, successful businesses often strive to provide excellent customer services. This underlying business goal can be translated into many different policies defining how network resources are to be used. One example of such a policy would be to prevent or limit access to non-business critical applications when performance of business critical applications is degraded beyond a threshold point. Another example would be to use QoS techniques to provide a guaranteed or high level of service to e-commerce applications. Yet another example would be to dynamically increase the network bandwidth allocated to a networked computer whenever it is accessed by a customer. Also, bandwidth for various applications might be restricted during times when there is heavy use of network resources by customers.

Control points 72 would access these policies and provide policy data to agents 70. Agents 70 and control points 72 would communicate with each other and monitor the network to determine how many customers were accessing the network, what computers the customer(s) were accessing, and what applications were being accessed by the customers. Once the triggering conditions were detected, the agents and control points would interact to re-allocate bandwidth, provide specified service levels, block or restrict various non-customer activities, etc. Another example of policy-based management would be to define an optimum specification of network resources or service levels for particular types of network tasks. The particular policies would direct the management entities to determine whether the particular task was permitted, and if permitted, the management entities would interact to ensure that the desired level of resources was provided to accomplish the task. If the optimum resources were not available, the applicable policies could further specify that the requested task be blocked, and that the requesting user be provided with an informative message detailing the reason why the request was denied. Alternatively, the policies could specify that the user be provided with various options, such as proceeding with the requested task, but with sub- optimal resources, or waiting to perform the task until a later time. For example, continuous media applications such as IP telephony have certain bandwidth requirements for optimum performance, and are particularly sensitive to network jitter and delay. Policies could be written to specify a desired level of service, including bandwidth requirements and threshold levels for jitter and delay, for client computers attempting to run IP telephony applications. The policies would further direct the agents and control modules to attempt to provide the specified level of service. Security checking could also be included to ensure that the particular user or client computer was permitted to run the application. In the event that the specified service level could not be provided, the requesting user could be provided with a message indicating that the resources for the request were not available. The user could also be offered various options, including proceeding with a sub- optimal level of service, placing a conventional telephone call, waiting to perform the task until a later time, etc.

The software, system and methods of the present description may be used to implement a wide variety of system policies. The policy rules and conditions may be based on any number of parameters, including IP source address, IP destination address, source port, destination port, protocol, application identity, user identity, device identity, URL, available device bandwidth, application profile, server profile, gateway identity, router identity, time- of-day, network congestion, network load, network population, available domain bandwidth and resource status, to name but a partial list. The actions taken when the policy conditions are satisfied can include blocking network access, adjusting service levels and/or bandwidth allocations for networked devices, blocking requests to particular URLs, diverting network requests away from overloaded or underperforming resources, redirecting network requests to alternate resources and gathering network statistics.

Some of the parameters listed above may be thought of as "client parameters," because they are normally evaluated by an agent monitoring a single networked client device. These include IP source address, IP destination address, source port, destination port, protocol, application identity, user identity, available device bandwidth and URL. Other parameters, such as application profile, server profile, gateway identity, router identity, time- of-day, network congestion, network load, network population, available domain bandwidth and resource status may be though of as "system parameters" because they pertain to shared resources, aggregate network conditions or require evaluation of data from multiple agent modules. Despite this, there is not a precise distinction between client parameters and system parameters. Certain parameters, such as time-of-day, may be considered either a client parameter or a system parameter, or both.

Policy-based network management, QoS implementation, and the other functions of the agents and control points depend on obtaining real-time information about the network. As will be discussed, certain described embodiments and implementations provide improvements over known policy-based QoS management solutions because of the enhanced ability to obtain detailed information about network conditions and the activity of networked devices. Many of the policy parameters and conditions discussed above are accessible due to the particular way the agent module embodiments may be coupled to the communications software of their associated devices. Also, as the above examples suggest, managing bandwidth and ensuring its availability for core applications is an increasingly important consideration in managing networks. Certain embodiments described herein provide for improved dynamic allocation of bandwidth and control of resource consumption in response to changing network conditions.

The ability of the systems described herein to flexibly deploy policy-based, QoS management solutions based on detailed information about network conditions has a number of significant benefits. These benefits include reducing frustration associated with using the network, reducing help calls to IT personnel, increasing productivity, lowering business costs associated with managing and maintaining enteφrise networks, and increased customer/user loyalty and satisfaction. Ultimately, the systems and methods ensure that network resources are used in a way that is consistent with underlying goals and objectives.

Implementation of policy-based QoS between the application and transport layers has another advantage. This allows support for encryption and other security implementations carried out using Virtual Private Networking (VPN) or IPSec protocol.

Referring now to Figs. 6-9, illustrative embodiments of the agent module will be more particularly described. The agent modules may monitor the status and activities of its associated client, server, pervasive computing device or other computing device; communicate this information to one or more control points; enforce system policies under the direction of the control points; and provide messages to network users and administrators concerning network conditions. Figs. 6-8 are conceptual depictions of networked computing devices, and show how the agent software may be associated with the networked devices relative to layered protocol software used by the devices for network communication.

As seen in Fig. 6, agent 70 is inteφosed between application program 122 and a communications protocol layer for providing end-to-end data transmission, such as transport layer 124 of communications protocol stack 92. Typically, the agent modules described herein may be used with network devices that employ layered communications software adhering to either the OSI or TCP/IP-based protocol models. Thus, agent 70 is depicted as "inteφosed," i.e. in a data path, between an application program and a transport protocol layer. However, it will be appreciated by those skilled in the art that the various agent module embodiments may be used with protocol software not adhering to either the OSI or TCP/IP models, but that nonetheless includes a protocol layer providing transport functionality, i.e. providing for end-to-end data transmission.

Because of the depicted position within the data path, agent 70 is able to monitor network traffic and obtain information that is not available by hooking into transport layer 124 or the layers below the transport layer. At the higher layers, the available data is richer and more detailed. Hooking into the stack at higher layers allows the network to become more "application-aware" than is possible when monitoring occurs at the transport and lower layers. The agent modules may be inteφosed at a variety of points between application program 122 and transport layer 124. Specifically, as shown in Figs. 7 and 8, agent 70 may be associated with a client computer so that it is adjacent an application programming interface (API) adapted to provide a standardized interface for application program 122 to access a local operating system (not shown) and communications stack 92. In Fig. 7, agent 70 is adjacent a winsock API 128 and inteφosed between application program 122 and the winsock interface. Fig. 8 shows an alternate configuration, in which agent 70 again hooks into a socket object, such as API 128, but downstream of the socket interface. With either configuration, the agent is inteφosed between the application and transport layer 124 of communications stack 92, and is adapted to directly monitor data received by or sent from the winsock interface.

As shown in Fig. 8, agent 70 may be configured to hook into lower layers of communications stack 92. This allows the agent to accurately monitor network traffic volumes by providing a correction mechanism to account for data compression or encryption occurring at protocol layers below transport layer 124. For example, if compression or encryption occurs within transport layer 124, monitoring at a point above the transport layer would yield an inaccurate measure of the network traffic associated with the computing device. Hooking into lower layers with agent 70 allows network traffic to be accurately measured in the event that compression, encryption or other data processing that qualitatively or quantitatively affects network traffic occurs at lower protocol layers. An embodiment of the agent module is depicted in Fig. 9. As shown, agent 70 may include a redirector module 130, a traffic control module 132, an administrator module 134, a DNS module 136, a popapp module 138, a message broker module 140, a system services module 142, and a popapp 144. Redirector module 130 intercepts winsock API calls made by applications running on networked devices such as the client computers depicted in Figs. 2 and 3. Redirector module 130 then hands these calls to one or more of the other agent components for processing. As discussed with reference to Figs. 6-8, redirector module is positioned to allow the agent to monitor data at a data transmission point between an application program running on the device and the transport layer of the communications stack. Depending on the configuration of the agent and control point, the intercepted winsock calls may be rejected, changed, or passed on by agent 70.

Traffic control module 132 implements QoS and system policies and assists in monitoring network conditions. Traffic control module 132 implements QoS methods by controlling the network traffic flow between applications running on the agent device and the network link. The traffic flow is controlled to deliver a specified network service level, which may include specifications of bandwidth, data throughput, jitter, delay and data loss.

To provide the specified network service level, traffic control module 132 may maintain a queue or plurality of queues. When data is sent from the client to the network, or from the network to the client, redirector module 130 intercepts the data, and traffic module 132 places the individual units of data in the appropriate queue. The control points may be configured to periodically provide traffic control commands, which may include the QoS parameters and service specifications discussed above. In response, traffic control module 132 controls the passing of data into, through or out of the queues in order to provide the specified service level.

More specifically, the outgoing traffic rate may be controlled using a plurality of priority-based transmission queues, such as transmission queues 132a. When an application or process is invoked by a computing device with which agent 70 is associated, a priority level is assigned to the application, based on centrally defined policies and priority data supplied by the control point. Specifically, as will be discussed, the control points maintain user profiles, applications profiles and network resource profiles. These profiles include priority data which is provided to the agents.

Transmission queues 132a may be configured to release data for transmission to the network at regular intervals. Using the parameters specified in traffic control commands issued by a control point, traffic module 132 calculates how much data can be released from the transmission queues in a particular interval. For example, if the specified average traffic rate is lOOKBps and the queue release interval is 1ms, then the total amount of data that the queues can release in a given interval is 100 bits. The relative priorities of the queues containing data to be transmitted determine how much of the allotment may be released by each individual queue. For example, assuming there are only two queues, Ql and Q2, that have data queued for transmission, Ql will be permitted to transmit 66.66% of the overall allotted interval release if its priority is twice that of Q2. Q2 would only be permitted to release 33.33% of the allotment. If their priorities were equal, each queue would be permitted to release 50% of the interval allotment for forwarding to the network link.

If waiting data is packaged into units that are larger than the amount a given queue is permitted to release, the queue accumulates "credits" for intervals in which it does not release any waiting data. When enough credits are accumulated, the waiting message is released for forwarding to the network.

Similarly, to control the rate at which network traffic is received, traffic control module 132 may be configured to maintain a plurality of separate receive queues, such as receive queues 132b. In addition to the methods discussed above, various other methods may be employed to control the rate at which network traffic is sent and received by the queues. Also, the behavior of the transmit and receive queues may be controlled through various methods to control jitter, delay, loss and response time for network connections.

The transmit and receive queues may also be configured to detect network conditions such as congestion and slow responding applications or servers. For example, for each application, transmitted packets or other data units may be timestamped when passed out of a transmit queue. When corresponding packets are received for a particular application, the receive and send times may be compared to detect network congestion and/or slow response times for various target resources. This information may be reported to the control points and shared with other agents within the domain. The response time and other performance information obtained by comparing transmit and receive times may also be used to compile and maintain statistics regarding various network resources.

Using this detection and reporting mechanism, a control point may be configured to reduce network loads by instructing traffic control module 132 to close low priority sessions and block additional sessions whenever heavy network congestion is reported by one of the agents. In conjunction, as will be explained, popapp 138 module may provide a message to the user explaining why sessions are being closed. In addition to closing the existing sessions, the control point may be configured to instruct the agents to block any further sessions. This action may also be accompanied by a user message in response to attempts to launch a new application or network process. When the network load is reduced, the control point will send : a message to the agents allowing sessions.

In addition to identifying congestion and slow response times, traffic control module 132 may be more generally configured to aid in identifying downed or under-performing network resources. When a connection to a target resource fails, traffic module 132 notifies popapp modules 138, which in turn launches an executable to perform a root-cause analysis of the problem. Agent 70 then provides the control point with a message identifying the resource and its status, if possible.

In addition, when a connection fails, popapp module 138 may be configured to provide a message to the user, including an option to initiate an autoconnect routine targeting the unavailable resource. Enabling autoconnect causes the agent to periodically retry the unavailable resource. This feature may be disabled, if desired, to allow the control point to assume responsibility for determining when the resource becomes available again. As will be later discussed, the described system may be configured so that the control modules assume responsibility for monitoring unavailable resources in order to minimize unnecessary network traffic. As discussed below, various agent components also monitor network conditions and resource usage for the puφose of compiling statistics. An additional function of traffic control module 132 is to aid in performing these functions by providing information to other agent components regarding accessed resources, including resource performance and frequency of access. As suggested in the above discussion of traffic control module 132, popapp module 138 stores and is responsible for launching a variety of small application modules such as application 144, known as popapps, to perform various operations and enhance the functioning of the described system. Popapps detect and diagnose network conditions such as downed resources, provide specific messages to users and IT personnel regarding errors and network conditions, and interface with other information management, reporting or operational support systems, such as policy managers, service level managers, and network and system management platforms. Popapps may be customized to add features to existing products, to tailor products for specific customer needs, and to integrate the software, systems and methods with technology supplied by other vendors.

Administrator module 134 interacts with various other agent modules, maintains and provides network statistics, and provides an interface for centrally configuring agents and other components of the system. With regard to agent configuration, administrator module 134 interfaces with configuration utility 106 (shown in Figs. 5 and 13 - 16), in order to configure various agent parameters. Administrator module 134 also serves as a repository for local reporting and statistics information to be communicated to the control points. Based on information obtained by other agent modules, administrator module 134 maintains local information regarding accessed servers, DNS servers, gateways, routers, switches, applications and other resources. This information is communicated on request to the control point, and may be used for network planning or to dynamically alter the behavior of agents. In addition, administrator module 134 stores system policies and/or components of policies, and provides policy data to various agent components as needed to implement and enforce the policies. Administrator module 134 also includes support for interfacing the described software and systems with standardized network management protocols and platforms. DNS module 136 provides the agent with configurable address resolving services.

DNS module 136 may include a local cache of DNS information, and may be configured to first resolve address requests using this local cache. If the request cannot be resolved locally, the request is submitted to a control point, which resolves the address with its own cache, provided the address is in the control point cache and the user has permission to access the address. If the request cannot be resolved with the control point cache, the connected control point submits the request to a DNS server for resolution. If the address is still not resolved at this point, the control point sends a message to the agent, and the agent then submits the request directly to its own DNS server for resolution.

DNS module 136 also monitors address requests and shares the content of the requests with administrator module 134. The requests are locally compiled and ultimately provided to the control points, which maintain dynamically updated lists of the most popular DNS servers, i addition, DNS module 136 is adapted to interact with control point 72 in order to redirect address resolving requests and other network requests to alternate targets, if necessary. Message broker module 140 creates and maintains connections to the one or more control points with which the agent interacts. The various agent components use the message broker module to communicate with each other and with a connected control point. Message broker module 140 includes message creator and message dispatcher processes for creating and sending messages to the control points. The message creator process includes member functions, which create control point messages by receiving message contents as parameters and encoding the contents in a standard network format. The message creator process also includes a member function to decode the messages received from the control point and return the contents in a format usable by the various components of the agent.

After encoding by the creator process, control point messages are added to a transmission queue and extracted by the message dispatcher function for transmission to the control point. Messages extracted from the queue are sent to the agent's active control point.

In addition, the dispatcher may be configured to ensure delivery of the message using a sequence numbering scheme or other error detection and recovery methods.

Messages and communications from an agent to a control point are made using a unicast addressing mechanism. Communications from the control point to an agent or agents may be made using unicast or a multicast addressing scheme. When configured for multicast operation, the control point and agents may be set to revert to unicast to allow for communication with devices that do not support IP multicast.

Once a connection with a control point is established, message broker module 140 monitors the status of the connection and switches over to a backup control point upon detecting a connection failure. If both the active and backup connections are not active, network traffic is passed on transparently.

System services module 142 provides various support functions to the other agent components. First, system services module maintains dynamic lists of user profiles, server profiles, DNS server profiles, control point connections and other data. The system services module also provides a tracing capability for debugging, and timer services for use by other agent components. System services module may also be configured with a library of APIs to interface the agent with the operating systems and other components of the device that the agent is associated with. Referring now to Figs. 10-12, control point 72 and its functions will be more particularly described. As seen in Fig. 10, control point may include a traffic module 160, a server profile module 162, a DNS server profile module 164, a gateway profile module 166, an administrator module 168, a message broker module 170, a popapp interface 172 and a popapp 174.

Control point traffic module 160 implements policy-based, QoS techniques by coordinating the service-level enforcement activities of the agents. As part of this function, traffic module 160 dynamically allocates bandwidth among the agents in its domain by regularly obtaining allocation data from the agents, calculating bandwidth allocations for each agent based on this data, and communicating the calculated allocations to the agents for enforcement. For example, control point 72 can be configured to recalculate bandwidth allocations every five seconds. During each cycle, between re-allocation, the agents restrict bandwidth usage by their associated devices to the allocated amount and monitor the amount of bandwidth actually used. At the end of the cycle, each agent reports the bandwidth usage and other allocation data to the control point to be used in re-allocating bandwidth.

During re-allocation, traffic module 160 divides the total bandwidth available for the upcoming cycle among the agents within the domain according to the priority data reported by the agents. The result is a configured bandwidth CB particular to each individual agent, corresponding to that agent's fair share of the available bandwidth. The priorities and configured bandwidths are a function of system policies, and may be based on a wide variety of parameters, including application identity, user identity, device identity, source address, destination address, source port, destination port, protocol, URL, time of day, network load, network population, and virtually any other parameter concerning network resources that can be communicated to, or obtained by the control point. The detail and specificity of client-side parameters that may be supplied to the control point is greatly enhanced by the position of agent redirector module 130 relative to the layered communications protocol stack. The high position within the stack allows bandwidth allocation and, more generally, policy implementation, to be performed based on very specific triggering criteria. This may greatly enhance the flexibility and power of the described software, systems and methods. The priority data reported by the agents may include priority data associated with multiple application programs running on a single networked device. In such a situation, the associated agent may be configured to report an "effective application priority," which is a function of the individual application priorities. For example, if device A were running two application programs and device B were running a single application program, device A's effective application priority would be twice that of device B, assuming that the individual priorities of all three applications were the same. The reported priority data for a device running multiple application programs may be further refined by weighting the reported priority based on the relative degree of activity for each application program. Thus, in the previous example, if one of the applications running on device A was dormant or idle, the contribution of that application to the effective priority of device A would be discounted such that, in the end, device A and device B would have nearly the same effective priority. To determine effective application priority using this weighted method, the relative degree of activity for an application may be measured in terms of bandwidth usage, transmitted packets, or any other activity-indicating criteria. In addition to priority data, each agent may be configured to report the amount of bandwidth UB used by its associated device during the prior period, as discussed above. Data is also available for each device's allocated bandwidth AB for the previous cycle. Traffic module 160 may compare configured bandwidth CB, allocated bandwidth AB or utilized bandwidth UB for each device, or any combination of those three parameters to determine the allocations for the upcoming cycle. To summarize the three parameters, UB is the amount the networked device used in the prior cycle, AB is the maximum amount they were allowed to use, and CB specifies the device's "fair share" of available bandwidth for the upcoming cycle.

Both utilized bandwidth UB and allocated bandwidth AB may be greater than, equal to, or less than configured bandwidth CB. This may happen, for example, when there are a number of networked devices using less than their configured share CB. To efficiently utilize the available bandwidth, these unused amounts are allocated to devices requesting additional bandwidth, with the result being that some devices are allocated amount AB that exceeds their configured fair share CB. Though AB and UB may exceed CB, utilized bandwidth UB cannot normally exceed allocated bandwidth AB, because the agent traffic control module enforces the allocation.

Any number of processing algorithms may be used to compare CB, AB and UB for each agent in order to calculate a new allocation, however there are some general principles which are often employed. For example, when bandwidth is taken away from devices, it is often desirable to first reduce allocations for devices that will be least affected by the downward adjustment. Thus, traffic module 160 may be configured to first reduce allocations of clients or other devices where the associated agent reports bandwidth usage UB below the allocated amount AB. Presumably, these devices won't be affected if their allocation is reduced. Generally, traffic module 160 should not reduce any other allocation until all the unused allocations, or portions of allocations, have been reduced. The traffic module may be configured to then reduce allocations that are particularly high, or make adjustments according to some other criteria.

Traffic module 160 may also be configured so that when bandwidth becomes available, the newly-available bandwidth is provisioned according to generalized preferences. For example, the traffic module can be configured to provide suφlus bandwidth first to agents that have low allocations and that are requesting additional bandwidth. After these requests are satisfied, suφlus bandwidth may be apportioned according to priorities or other criteria.

Figs. 11A, 11B, 11C and 11D depict examples of various methods that may be implemented by traffic module 160 to dynamically allocate bandwidth. Fig. 11A depicts a process by which traffic module 160 determines whether any adjustments to bandwidth allocations AB are necessary. Allocated bandwidths AB for certain agents are adjusted in at least the following circumstances. First, as seen in steps S4 and S10, certain allocated bandwidths AB are modified if the sum of all the allocated bandwidths ABtotal exceeds the sum of the configured bandwidths CBtotal. This situation may occur where, for some reason, a certain portion of the total bandwidth available to the agents in a previous cycle becomes unavailable, perhaps because it has been reserved for another puφose. In such a circumstance, it is important to reduce certain allocations AB to prevent the total allocations from exceeding the total bandwidth available during the upcoming cycle.

Second, if there are any agents for which AB < CB and UB = AB, the allocation for those agents is modified, as seen in steps S6 and S10. The allocations for any such agent are typically increased. In this situation, an agent has an allocation AB that is less than their configured bandwidth CB, i.e. their existing allocation is less than their fair share of the bandwidth that will be available in the upcoming cycle. Also, the reported usage UB for the prior cycle is at or near the enforced allocation AB, and it can thus be assumed that more bandwidth would be consumed by the associated device if its allocation AB were increased.

Third, if there are any agents reporting bandwidth usage UB that is less than their allocation AB, as determined at step S8, then the allocation AB for such an agent is reduced for the upcoming period to free up the unused bandwidth. Steps S4, S6 and S8 may be performed in any suitable order. Collectively, these three steps ensure that certain bandwidth allocations are modified, i.e. increased or reduced, if one or more of the following three conditions are true: (1) ABtotal > CBtotal, (2) AB < CB and UB ≡ AB for any agent, or (3) UB < AB for any agent. If none of these are true, the allocations AB from the prior period are not adjusted. Traffic module 160 modifies allocations AB as necessary at step S10. After all necessary modifications are made, the control point communicates the new allocations to the agents for enforcement during the upcoming cycle. Fig. 1 IB depicts re-allocation of bandwidth to ensure that total allocations AB do not exceed the total bandwidth available for the upcoming cycle. At step SI 8, traffic module 160 has determined that the sum of allocations AB from the prior period exceed the available bandwidth for the upcoming period, i.e. ABtotal > CBtotal. In this situation, certain allocations AB must be reduced. As seen in steps S20 and S22, traffic module 160 may be configured to first reduce allocations of agents that report bandwidth usage levels below their allocated amounts, i.e. UB < AB for a particular agent. These agents are not using a portion of their allocations, and thus are unaffected or only minimally affected when the unused portion of the allocation is removed. At step S20, the traffic module first determines whether there are any such agents. At step S22, the allocations AB for some or all of these agents are reduced. These reductions may be gradual, or the entire unused portion of the allocation may be removed at once.

After any and all unused allocation portions have been removed, it is possible that further reductions may be required to appropriately reduce the overall allocations ABtotal. As seen in step S24, further reductions are taken from agents with existing allocations AB that are greater than configured bandwidth CB, i.e. AB > CB. In contrast to step S22, where allocations were reduced due to unused bandwidth, bandwidth is removed at step S24 from devices with existing allocations that exceed the calculated "fair share" for the upcoming cycle. As seen at step S26, the reductions taken at steps S22 and S24 may be performed until the total allocations ABtotal are less than or equal to the total available bandwidth CBtotal for the upcoming cycle.

Fig. 11C depicts a method for increasing the allocation of certain agents. As discussed with reference to Fig. 11 A, where AB < CB and UB ≡ AB for any agent, the allocation AB for such an agent should be increased. The existence of this circumstance has been determined at step S40. To provide these agents with additional bandwidth, the allocations for certain other agents typically need to be reduced. Similar to steps S20 and S22 of Fig. 1 IB, unutilized bandwidth is first identified and removed (steps S42 and S44). Again, the control point may be configured to vary the rate at which unused allocation portions are removed. If reported data does not reflect unutilized bandwidth, traffic module 160 may be configured to then reduce allocations for agents having an allocation AB higher than their respective configured share CB, as seen in step S46. The bandwidth recovered in steps S44 and S46 is then provided to the agents requesting additional bandwidth. Any number of methods may be used to provision the recovered bandwidth. For example, preference may be given to agents reporting the largest discrepancy between their allocation AB and their configured share CB. Alternatively, preferences may be based on application identity, user identity, priority data, other client or system parameters, or any other suitable criteria.

Fig. 1 ID depicts a general method for reallocating unused bandwidth. At step S60, it has been determined that certain allocations AB are not being fully used by the respective agents, i.e. UB < AB for at least one agent. At step S62, the allocations AB for these agents are reduced. As with the reductions and modifications described with reference to Figs. 11A, 1 IB and 11C, the rate of the adjustment may be varied through configuration changes to the control point. For example, it may be desired that only a fraction of unused bandwidth be removed during a single reallocation cycle. Alternatively, the entire unused portion may be removed and reallocated during the reallocation cycle.

In step S64 of Fig. 1 ID, the recovered amounts are provisioned as necessary. The recovered bandwidth may be used to eliminate a discrepancy between the total allocations ABtotal and the available bandwidth, as in Fig. 11B, or to increase allocations of agents who are requesting additional bandwidth and have relatively low allocations, as in Fig. l lC. hi addition, if there is enough bandwidth recovered, allocations may be increased for agents requesting additional bandwidth, i.e. UB s AB, even where the current allocation AB for such an agent is fairly high, e.g. AB > CB. As with the methods depicted in Figs. 11B and 11C, the recovered bandwidth may be reallocated using a variety of methods and according to any suitable criteria. As indicated, traffic module 160 can be configured to vary the rate at which the above allocation adjustments are made. For example, assume that a particular device is allocated 64 KBps (AB) and reports usage during the prior cycle of 62 KBps (UB). Traffic module 160 cannot determine how much additional bandwidth the device would use. Thus, if the allocation were dramatically increased, say doubled, it is possible that a significant portion of the increase would go unused. However, because the device is using an amount roughly equal to the enforced allocation AB, it can be assumed that the device would use more if the allocation were increased. Thus, it is often preferable to provide small, incremental increases. The amount of these incremental adjustments and the rate at which they are made may be configured with the configuration utility, as will be discussed with reference to Fig. 16. If the device consumes the additional amounts, successive increases can be provided if additional bandwidth is available. i addition, the bandwidth allocations and calculations may be performed separately for the transmit and receive rates for the networked devices, hi other words, the methods described with reference to Figs. 11 A - 1 ID may be used to calculate a transmit allocation for a particular device, as well as a separate receive allocation. Alternatively, the calculations may be combined to yield an overall bandwidth allocation.

Server profile module 162, DNS server profile module 164, gateway profile module 166 and administrator module 168 all interact with the agents to monitor the status of network resources. More specifically, Fig. 12 provides an illustrative example of how the control points and agents may be configured to monitor the status of resources on the network. The monitored resource(s) may be a server, a DNS server, a router, gateway, switch, application, etc. At step SI 00, a resource status change has occurred. For example, a server has gone down, traffic through a router has dramatically increased, a particular application is unavailable, or the performance of a particular gateway has degraded beyond a predetermined threshold specified in a system policy.

At step SI 02, a networked device attempts to access the resource or otherwise engages in activity on the network involving the particular resource. If the accessing or requesting device is an agent, an executable spawned by popapp module 138 (Fig. 9) analyzes the resource, and reports the identity and status of the resource to the control point connected to the agent, as indicated at step S104. Launch of the popapp may be triggered by connection errors, or by triggering criteria specified in system policies. For example, system policies may include performance benchmarks for various network resources, and may further specify that popapp analysis is to be performed when resource performance deviates by a certain amount from the established benchmark. In addition, the control points may similarly be configured to launch popapps to analyze network resources.

Once the control point obtains the status information, the control point reports the information to all of the agents in its domain, and instructs the agents how to handle further client requests involving the resource, as indicated at steps S108 and SI 10. In the event that the target resource is down, undeφerforming or otherwise unavailable, the instructions given to the agents will depend on whether an alternate resource is available. The control point stores dynamically updated lists of alternate available resources. If an alternate resource is available, the instructions provided to the agent may include an instruction to transparently redirect the request to an alternate resource, as shown in step SI 08. For example, if the control point knows of a server that mirrors the data of another server that has gone down, client requests to the down server can simply be redirected to the mirror server. Alternatively, if no alternate resource is available, the agent can be instructed to provide a user message in the event of an access attempt, as seen in step SI 10. The messaging function is handled by agent popapp module 138. In addition, popapp functionality may be employed by the control point to report status information to other control points and management platforms supplied by other vendors. In addition, messages concerning resource status or network conditions may be provided via email or paging to IT personnel.

Still referring to Fig. 12, the control point may be configured to assume responsibility for tracking the status of the resource in order to determine when it again becomes available, as shown in step SI 12. A slow polling technique is used to minimize unnecessary traffic on the network. During the interval in which the resource is unavailable, the agents either redirect requests to the resources or provide error messages, based on the instructions provided by the control point, as shown in step SI 16. Once the control point determines that the resource is again available, the control point shares this information with the agents and disables the instructions provided in steps S108 and SI 10, as shown in step SI 18.

This method of tracking and monitoring resource status has important advantages. First, it reduces unnecessary and frustrating access attempts to unavailable resources. Instead of repeatedly attempting to perform a task, a user's requests are redirected so that the request can be serviced successfully, or the user is provided with information about why the attempt(s) was unsuccessful. With this information in hand, the user is less likely to generate wasteful network traffic with repeated access attempts in a short period of time, hi addition, network traffic is also reduced by having only one entity, usually a control point, assume responsibility for monitoring a resource that is unavailable.

In addition to assisting these resource monitoring functions, server profile module 162 maintains a dynamically updated list of the servers accessed by agents within its domain. The server statistics may be retrieved using the configuration utility, or with a variety of other existing management platforms. The server statistics may be used for network planning, or may be implemented into various system policies for dynamic enforcement by the agents and control points. For example, the control points and agents can be configured to divert traffic from heavily used servers or other resources.

DNS module 164 also performs certain particularized functions in addition to aiding the resource monitoring and tracking described with reference to Fig. 12. Specifically, the DNS module maintains a local DNS cache for efficient local address resolution. As discussed with reference to agent DNS module 136, the agents and control points interact to resolve address requests, and may be configured to resolve addresses by first referencing local DNS data maintained by the agents and/or control points. Similar to server profile module 162, DNS module 164 also maintains statistics for use in network planning and dynamic system policies.

In addition to the functions described above, administrator module 168 maintains control point configuration parameters and distributes this information to the agents within the domain. Similar to the server, DNS and gateway modules, administrator module 168 also aids in collecting and maintaining statistical data concerning network resources. In addition, administrator module 168 retrieves policy data from centralized policy repositories, and stores the policy data locally for use by the control points and agents in enforcing system policies. Control point 72 also includes a synchronization interface (not shown) for synchronizing information among multiple control points within the same domain.

Message broker module 170 performs various functions to enable the control point to communicate with the agents. Similar to agent message broker module 140, message broker module 170 includes message creator and message dispatcher processes. The message creator process includes member functions that receive message contents as parameters and return encoded messages for transmission to agents and other network entities. Functions for decoding received messages are also included with the message creator process. The dispatcher process transmits messages and ensures reliable delivery through a retry mechanism and error detection and recovery methods. Referring now to Figs. 13 - 16, both the agents and control points may be configured using configuration utility 106. Typically, configuration utility 106 is a platform-independent application that provides a graphical user interface for centrally managing configuration information for the control points and agents. To configure the control points and agents, the configuration utility interface with administrator module 134 of agent 70 and with administrator module 168 of control point 72. Alternatively, configuration utility 106 may interface with administrator module 168 of control point 72, and the control point in turn may interface with administrator module 134 of agent 70.

Fig. 13 depicts a main configuration screen 188 of configuration utility 106. As indicated, main configuration screen 188 can be used to view various managed objects, including users, applications, control points, agents and other network entities and resources. For example, screen frame 190 on the left side of the main configuration screen 188 may be used to present an expandable representation of the control points that are configured for the network.

When a particular control point is selected in the main configuration screen 188, various settings for the control point may be configured. For example, the name of the control point may be edited, agents and other entities may be added to the control point's domain, and the control point may be designated as a secondary connection for particular agents or groups of agents. In addition, the system administrator may specify the total bandwidth available to agents within the control point's domain for transmitting and receiving, as shown in Fig. 14. This bandwidth specification will affect the configured bandwidths CB and allocated bandwidths AB discussed with reference to control point traffic module 160 and the method depicted in Fig. 11.

Configuration utility 106 also provides for configuration of various settings relating to users, applications and resources associated with a particular control point. For example, users may be grouped together for collective treatment, lists of prohibited URLs may be specified for particular users or groups of users, and priorities for applications may be specified, as shown in Fig. 15. Priorities may also be assigned to users or groups of users. As discussed above, this priority data plays a role in determining bandwidth allocations for the agents and their associated devices. In addition, optimum and minimum performance levels may be established for applications or other tasks using network resources. Referring again to the IP telephony example discussed above, the configuration utility may be used to specify a minimum threshold performance level for a networked device running the IP telephony application. This performance level may be specified in terms of QoS performance parameters such as bandwidth, throughput, jitter, delay and loss. The agent module associated with the networked device would then monitor the network traffic associated with the IP telephony application to ensure that performance was above the minimum threshold. If the minimum level was not met, the control points and agents could interact to reallocate resources and provide the specified minimum service level. Similarly, an optimum service level may be specified for various network applications and tasks. More generally, configuration utility 106 may be configured to manage system policies by providing functionality for authoring, maintaining and storing system policies, and for managing retrieval of system policies from other locations on a distributed network, such as a dedicated policy server.

Referring now to Fig. 16, the configuration of various other control point and agent parameters will be discussed. As seen in the figure, configuration utility 106 may be used to configure the interval at which resource reallocation is performed. For example, the default interval for recalculating bandwidth allocations is 5000 milliseconds, or 5 seconds. Also, as discussed above, the rate at which resource allocation occurs may be specified in order to prevent overcompensation, unnecessary adjustments to allocations, and inefficient reconfigurations. Specifically, the percentage of over-utilized bandwidth that is removed from a client device and reallocated elsewhere may be specified with the configuration utility, as seen in Fig. 16. In addition, the rate at which agents provide feedback to the control points regarding network conditions or activities of their associated devices may be configured.

In many of the examples discussed above, the systems and methods are implemented architecturally in two tiers. The first tier may include one or more control modules, such as control points 72. Because the control points control and coordinate operation of agent modules 70, the control points may be referred to as "upstream" or "overlying" components, relative to the agent modules that they control. By contrast, the agents, which form the second tier of the system, may be referred to as "downstream" or "underlying" components, relative to the control points they are controlled by.

It will be appreciated that the systems and methods described herein are extremely flexible and scalable, and may be applied to networks of widely varying size, hi some settings, scaling is achieved by extending hierarchical implementations to three or more tiers. Indeed, a very large architectural model may be built by extending the two-tier example above on the upstream and/or downstream side.

Such a model can be used to deliver consistent application performance in a very large and complex network configuration. An upstream component may control one or more downstream components. An upstream component can also be configured to be a downstream component of some other controlling upstream entity. Similarly, a downstream component may be configured to be an upstream component to control its downstream delegates.

Fig. 17 depicts an exemplary multi-tier implementation. In this example, the Tier 1 component is upstream relative to three components in Tier 2. Because the Tier 2 components underlie and are controlled by the Tier 1 component, the Tier 2 components are downstream components relative to the Tier 1 component. As indicated, the components within Tiers 2, 3 and 4 may be configured to function as both downstream and upstream components.

The agent modules and control points described herein may be implemented at any level within a multi-tier environment such as that shown in Fig. 17. For example, the Tier 4 component may include a control module 72, as shown, that is configured to control agent modules 70 that may be running on the underlying Tier 5 components.

Hierarchically, the structure of Fig. 17 typically is implemented in a one-to-many configuration moving downstream. In other words, an upstream component may control multiple downstream components in the tier immediately below it, and those downstream components may control multiple components in further downstream tiers. However, a given downstream component typically reports to and is controlled by only a single upstream component. It should be appreciated that a wide variety of configurations are possible, with any desired number of tiers and components or groupings of components within the tiers.

In multi-tier environments such as that described above, it will often be desirable to provide mechanisms for centrally defining and distributing policies relating to management of bandwidth and other resources. Fig. 18 depicts an exemplary networking environment employing such mechanisms. In the example, various branch offices are connected to a centralized data center 200 via network 202. The system may be configured to enable an administrator to define enteφrise wide policies on a central server such as an Enteφrise Policy Server (EPS) 204. In such an environment, the control point modules described herein may be implemented in connection with a Controlled Location Policy Server (CLPS) 206, as shown in the exemplary branch office 208 at the bottom of Fig. 18.

Typically, a given CLPS 206 retrieves policies pertinent to its branch office from its controlling EPS 204. The CLPS, which may include a control module 72, then distributes pertinent policies to the relevant agent modules 70 within its domain. The retrieved policies are then distributed to the various agents that are controlled by that control point module. This policy definition and distribution scheme may easily be adapted and scaled to manage widely varying enteφrise configurations. The distributed policies may, among other things, be used to facilitate the bandwidth management techniques described herein. In addition to or instead of the previously described bandwidth management features, bandwidth management may be effected using tiered implementations such as that described above. Fig. 19 depicts an exemplary system 300 for managing bandwidth on network link 302.

As shown, system 300 may include various components referred to as bandwidth managers, which may be implemented within the control modules (e.g., control points 72) and agent modules (e.g., agent modules 70) described herein. For example, the depicted system includes a controlled location bandwidth manager (CL-BWMGR) 304, which may be implemented within previously described control point 72, and/or within a Controlled Location Policy Server 206, such as shown in Fig. 18. In any case, CL-BWMGR 304 typically is implemented as a software program running on a computing device, such as a server (not shown), connected to network link 302.

The CL-BWMGR typically communicates with agent software (e.g., agent modules 70) running on computing devices connected to network link 302, so as to manage bandwidth usage by those devices. For example, as shown, plural computing devices 22, also referred to as agent computers, may be interconnected via network link 302. Each agent computer may be running an agent bandwidth manager (AG-BWMGR) 306, one or more application bandwidth managers (APP-BWMGR) 308, and one or more socket bandwidth managers (SOCK-BWMGR) 310. The bandwidth managers loaded on agent computer 22 typically are sub-components of agent module 70, discussed above. Within a given computing device 22 connected to network link 302, typically there is one AG-BWMGR 306, and a variable number of APP-BWMGRs 308 and SOCK-BWMGRs 310, depending on the applications and sockets being utilized. For example, when the bandwidth managers are implemented within an agent module 70, the agent module typically is adapted to launch an APP-BWMGR 308 for each application 320 running on computer 22, and a SOCK-BWMGR 310 for each socket 322 open by each application 320. As indicated, a given computing device may be running multiple applications 320, and a given application 320 may have multiple open sockets 322. Computing devices 22 transmit and receive data on network link 302 via transmit and receive control sections 324 and 326. Sections 324 and 326 may respectively include transmit and receive queues 328 and 330, as will be explained below.

In the depicted example, CL-BWMGR 304 manages overall bandwidth on network link 302, a given AG-BWMGR 306 manages bandwidth usage by its associated agent computer 22, a given APP-BWMGR 308 manages bandwidth usage by its associated application 320, and a given SOCK-BWMGR 310 manages bandwidth usage by its associated socket 322.

As indicated by the arrows interconnecting the different bandwidth managers, the bandwidth managers may be arranged in a hierarchical control configuration, in which an upstream component interacts with and/or controls one or more underlying downstream components. In the depicted example, the CL-BWMGR manages bandwidth usage on network link 302 by interacting with underlying AG-BWMGRs, so as to manage bandwidth usage by the particular downstream computing devices 22, applications 320 and sockets 322 that underlie the CL-BWMGR. It should be appreciated that the CL-BWMGR may manage bandwidth usage on any type of interconnection between computers. For example, in Fig. 2, a given remote network 14 could include a server computer running a CL-BWMGR 304 interconnected via a local network segment with plural client computers, where each client computer was loaded with an agent module 70. In such a case, the control computer could control the clients' bandwidth usage on not only the local network segment (e.g., an Ethernet- based LAN), but also on the connection through router 18 to public network 16. Referring again to Fig. 19, a given AG-BWMGR 306 manages bandwidth usage by its associated computing device 22 by interacting with underlying APP-BWMGRs, so as to manage bandwidth usage by the particular downstream applications 320 and sockets 322 that underlie the AG-BWMGR. Likewise, a given APP-BWMGR 308 manages bandwidth usage by its associated application 320 by interacting with underlying SOCK-BWMGRs, so as to manage bandwidth usage by the particular downstream sockets 322 that underlie the APP- BWMGR. As explained below, downstream components typically facilitate control by reporting certain data, such as bandwidth consumption, upstream to overlying upstream components.

In the depicted example, bandwidth may be managed by taking a bandwidth allocation existing at a particular hierarchical level, and sub-allocating the allocation for apportionment amongst downstream components. Where there are multiple downstream components, sub-allocation typically involves dividing the allocation into portions for each of the downstream components.

For example, CL-BWMGR 304 may have an allocation corresponding to the available bandwidth on network link 302. The available bandwidth may be configured according to a system policy specifying parameters for network link 302, or may be determined or configured through other methods. The available bandwidth on link 302 may be sub-allocated by CL-BWMGR 304 into one or more agent allocations, depending on the number of agent computers 22 underlying the CL-BWMGR. Each agent allocation represents the amount of bandwidth allotted to the respective agent computer. For example, if one hundred agent computers 22 are controlled by the CL-BWMGR, then the CL-BWMGR would typically sub-allocate available link bandwidth into one hundred individualized agent allocations for the respective underlying agent computers. The individualized agent allocation would be provided to the respective AG-BWMGRs at each agent computer 22 for enforcement and further sub-allocation.

Similarly, at each agent computer, the associated AG-BWMGR 306 may sub-allocate its agent allocation into individualized application allocations for each application 320 running on the agent computer. A given application allocation represents the amount of bandwidth allotted to the corresponding application. These application allocations typically are provided to the respective APP-BWMGRs for enforcement and further sub-allocation. Finally, at each application 320, the associated APP-BWMGR 308 may sub-allocate its application allocation into individual socket allocations for the sockets 322 that are open for that application. The individual socket allocations typically are provided to the respective SOCK-BWMGRs 310. Typically, at least some of the bandwidth managers are configured to communicate upstream in order to facilitate the various sub-allocations discussed above. Indeed, data may be provided upstream from each group of SOCK-BWMGRs 310 to their overlying APP- BWMGR 308, from each group of APP-BWMGRs 308 to their overlying AG-BWMGR 306, and from each group of AG-BWMGRs 306 to their overlying CL-BWMGR 304. The data provided upstream may be used to calculate sub-allocations, which are then sent back downstream for enforcement, and/or further sub-allocation. Often, it will be desirable that the data which is sent upstream pertain to socket activity, so as to efficiently adjust future allocations to take bandwidth from where it is less needed and provide it to where it is more needed. Indeed, the interaction between the various bandwidth managers typically is performed to efficiently allocate network bandwidth among the various bandwidth-consuming components. As indicated above, consuming components may include agent computers 22, applications 320 running on those computers, and sockets 322 open for those applications. One way in which the systems described herein may efficiently allocate bandwidth is by shifting bandwidth allocations toward high priority uses and away from uses of relatively lower priority. Another allocation criteria may be employed which involves providing future allocations based on consumption of past allocations. For example, allocations to a particular bandwidth manager (e.g., an application allocation provided from an AG-BWMGR 306 to an APP-BWMGR 308) may be reduced if past allocations were only partially consumed, or if there is a trend of diminished usage.

The interactions discussed above need not occur in any particular order, and may occur periodically or non-periodically, and/or at different rates for different levels. For example, CL-BWMGR 304 may hand out agent allocations for computers 22 at regular intervals, or only when a changed condition within the network is detected. Application allocations and socket allocations may be disseminated downstream periodically, but at different rates, hi any case, it will often be advantageous to repeatedly and dynamically update the allocations and sub-allocations, as will be discussed in more detail below.

Bandwidth management for a representative socket will now be described with reference to Fig. 20, which shows an agent computing device 22 coupled to other computers (not shown) via network link 302. Computing device 22 is running an application program 320 which communicates over network link 302 via a layered protocol stack (depicted in part at 92), as described with reference to earlier examples. Agent module 70 may be inteφosed between application program 320 and lower layers of stack 92, typically at a point where flow control may be achieved. Specifically, as described with reference to Figs. 6, 7 and 8, agent module 70 typically is positioned in a relatively high location within the protocol stack and interfaced with a socket object, to allow network traffic to be intercepted, or hooked into, at a point between application program 320 and transport layer 124 of stack 92. The agent module 70 depicted in Fig. 20 may include some or all of the components described with reference to Fig. 9. In particular, a redirector 130 (Fig. 9) may be employed to allow agent module to intercept, or hook into, the socket data flow between application 320 and network link 302.

For the depicted representative socket, traffic between application program 320 and network link 302 flows through transmit control section 324 and receive control section 326, which may include transmit queues 328 and receive queues 330, respectively. Data flows through control sections 324 and 326 may be monitored and controlled via SOCK-BWMGR 310 of agent module 70. As described above, SOCK-BWMGR 310 may interact with upstream bandwidth managers (not shown in Fig. 20) implemented within agent module 70. Specifically, socket allocations may be provided to SOCK-BWMGR 310, which controls transmit control section 324 and receive control section 326 to ensure that those allocations are enforced. Also, as indicated, SOCK-BWMGR 310 may provide feedback (e.g., concerning socket activity) to various upstream bandwidth managers. This feedback may be used to facilitate future allocations.

Socket allocations may be provided to SOCK-BWMGR 310 periodically at regular intervals. In such a case, the allocation typically is in the form of a number of bytes that may be transmitted, and/or a number of bytes that may be received, during a set interval (there may be separate transmit and receive allocations). During the interval, data passes through control sections 324 and 326, and SOCK-BWMGR 310 monitors the amount of data sent to, or received from, network link 302. Provided that the allocation for a given interval is not exceeded, SOCK-BWMGR 310 may simply passively monitor the traffic. However, once the allocation for a particular interval is used up, further transmission and reception is prevented, until a new allocation is obtained (e.g., during a succeeding interval). As indicated, queues 328 and 330 may be provided to hold overflow, until the socket is replenished with a new allocation, at which point the queued data may be transmitted and/or received. When an allocation is not exceeded, queues 328 and 330 typically are not used, such that transmitted or received data passes through sections 324 and/or 326 without any queuing of data, hi these implementations, queuing is triggered only when the socket allocation for a given interval is exceeded. hi many cases, it will be desirable that the upstream feedback for the socket (e.g., the feedback sent upstream by SOCK-BWMGR 310) include specification of how much bandwidth was consumed by the socket (e.g., how many bytes were sent or received). This consumption data may be used by upstream bandwidth managers in performing sub- allocations, as will be explained below.

In connection with the agent device management described herein, the term socket should be broadly understood to mean network conversations or connections carried out by applications above the transport protocol layer (e.g., layer 124 in Figs. 6, 7, 8 and 20). In most cases, these conversations or connections are characterized at least in part by the ability to implement flow control. Accordingly, it will be appreciated that the agent modules are not limited to any one particular standard that is employed within the upper layers of the protocol stack. Indeed, the agent modules described herein may hook into network conversations carried on by applications that do not use the pervasive Winsock standard.

Referring particularly to Fig. 21, exemplary agent computer 22 is shown as running applications 320a, 320b, and 320c. Application 320a employs the Winsock standard, and thus communicates with network link 302 via Winsock API 128 and lower layers of protocol stack (e.g., the layers shown at 340 and 342). Accordingly, agent module 70a may be inteφosed as shown relative to API 128 to hook into the network communications and carry out the monitoring and management functions described at length herein.

In contrast, applications 320b and 320c do not work with the Winsock standard. An example of such an application is an Active Directory application that employs the NetBios standard. For these non-winsock applications, an alternate agent module 70b may be provided. Similar to module 70a and the previously discussed agent module embodiments, agent module 70b is able to hook into network conversations above the transport layer. The module does not require winsock, though it is still active at a high protocol layer where flow control may be employed and where a rich variety of parameters may be accessed to enhance monitoring and policy-based management.

Returning to the discussion of bandwidth management, various different methodologies may be employed to provide downstream sub-allocations of bandwidth. Many of these methodologies employ a priority scheme, in which priorities may be assigned via network policies. Priority may be assigned based on a variety of parameters, including IP source address, IP destination address, source port, destination port, protocol, application identity, user identity, device identity, URL, available device bandwidth, application profile, server profile, gateway identity, router identity, time-of-day, network congestion, network load, network population, available domain bandwidth and status of network resources. Other parameters are possible. As discussed above, the position of the described agent modules relative to the protocol stack allows access to many different parameters. The ability to predicate bandwidth management on such varying criteria allows for increased control over bandwidth allocations, and thereby increases efficiency.

In typical implementations, the assigned priorities discussed above may be converted into or expressed in terms of a priority level that is specific to a given network "conversation," or connection, such as a socket data flow. Indeed, many of the sub-allocation schemes discussed below depend on socket priorities for individual sockets, h the policy- based systems described herein, the socket priorities typically are configured values derived from various parameters such as those listed above. For example, the configured priority for a given socket might be determined by a combination of user identity, the type of application being run, and the source address of the requested data, to name but one example.

Referring again to Fig. 19, various exemplary methods for sub-allocating bandwidth will now be discussed. Beginning with the sub-allocation of an application allocation into one or more socket allocations, as may be performed by APP-BWMGR 308, assume that there are k sockets 322 open for a given application 320. For a given socket s (where s = 1 through k), an effective socket priority ESP(s) may be calculated as follows:

E r-SPr.(s) = S ΓTPT-,(s x) x ^COnS = ^^aSt N(s) ^J x f _ractor , (l) alloc _last _N(s)

where SP(s) is the configured priority for the socket (e.g., as determined by network policies); cons_last_N(s) is the bandwidth consumed by the socket during the last N allocation cycles; and alloc_last_N(s) is the bandwidth allocated to the socket over the last N cycles. In addition, the result may be multiplied by a convenient factor to avoid handling of fractional amounts.

Then, the socket allocation SALLOC(s) for a given socket may be calculated as:

SALLOCis) = APALLOC ^SP^ , (2)

∑ESP s) s=l where APALLOC is the application allocation being apportioned by the APP-BWMGR among the various sockets; and ESP(s) is the effective priority of the socket. It should thus be appreciated that the apportionment is derived via a weighted average of the effective socket priorities.

Accordingly, in the above example, socket allocations will tend to be higher for sockets having higher priorities, and for sockets consuming larger percentages of prior allocations. In the above example, calculations are based on consumption data from multiple past allocation cycles (N cycles), though a single cycle calculation may be used. In some cases, multiple cycle calculations may smooth transitions and stabilize adjustment of allocations among the various sockets. Typically, the calculations described above are performed by the overlying APP-BWMGR based on bandwidth consumption data received from underlying SOCK-BWMGRs. Turning now to the sub-allocation of agent allocations into one or more application allocations, as may be performed by AG-BWMGR 306, assume that there are j applications

320 running on a given computer 22. First, an effective application priority may be calculated for each application. Using the above application as an example, the effective application priority EAPP may be calculated as follows: _π , _nn A cons last N(s) x SP(s) ~ _^

EAPP = J = ^{~ }} — x factor , (3) _s=ι alloc _last _N(s)

Again, a factor may be employed to avoid handling of fractions, or to otherwise facilitate processing. A similar calculation is performed for all of the other applications 320 underlying the AG-BWMGR. Then, similar to the socket sub-allocations above, the individual application allocations may be produced with a weighted average. Specifically, for a given application ap (where ap = 1 through j), an application allocation APPALLOC(ap) may be calculated as follows:

APALLOC(ap) = AGALLOC x ^EAPP^^aP^ , ₍₄₎

∑EAPP(ap) ap=Λ where AGALLOC is the agent allocation being apportioned by the AG-BWMGR among the various applications; and EAPP(ap) is the effective priority of the application for which the allocation is being calculated. Typically, the calculations described above are performed by the overlying AG-BWMGR based on bandwidth consumption data received from underlying APP-BWMGRs 308. In the above example, application allocations will tend to be higher for applications having more sockets open, higher priority sockets, and sockets consuming greater percentages of prior socket allocations.

Turning now to the sub-allocation of link bandwidth (e.g., overall bandwidth available on network link 302) into one or more agent allocations, as may be performed by CL-BWMGR 304, assume that there are i agent computers 22 interconnected by network link 302. As with the above sub-allocations, an effective priority may be calculated and then bandwidth may be apportioned according to a weighted average of the effective priorities. Specifically, using the above agent as an example, the effective agent priority EAGP may be calculated by summing the underlying effective application priorities as follows:

EAGP = ∑EAPP(ap) (5) ap=\ A similar calculation may be performed for all of the other agent computers underlying the CL-BWMGR, in order to obtain all of the effective agent priorities, hi some cases, it may be desirable to weight the newly calculated value with the prior value to obtain the effective priority, in order to smooth adjustments and bandwidth reallocations among the agent computers, hi certain implementations, for example, it has proved advantageous to calculate the effective agent priority by blending the new value (derived from above equation) with the most recent value in a 60-40 ratio. Any other desirable weighting may be employed, and/or other methods may be used to smooth allocation transitions. In any case, the sub- allocation to the agent computers may be effected with a weighted average:

AGALLOC(ag) = BWCL x ^^AGP(a^ , ₍₆₎

∑EAGP(ag) αg=l where AGALLOC(ag) is the agent allocation for a given agent computer ag (ag = 1 through i); BWCL is the bandwidth available to be allocated among all the agent computers; and EAGP(ag) is the effective agent priority for the given agent computer. hi the above examples, the activity of individual sockets typically is communicated upstream (e.g., from a SOCK-BWMGR to an overlying APP-BWMGR, to an overlying AG- BWMGR, and to an overlying CL-BWMGR), and has an effect on future allocations and sub- allocations throughout the system. Consumption at a particular socket may affect future allocations to that socket and other sockets on the same application. In addition, because the consumption data is communicated upstream and used in other allocations and sub- allocations, the same socket can potentially affect future allocations to all sockets, applications and agent computers within the system, hi certain implementations this feedback effect can greatly increase the efficiency of bandwidth management.

Those skilled in the art will appreciate that allocations and sub-allocations may be performed separately for transmitted data and received data. In addition, the upstream feedback and downstream sub-allocations may occur at any desired frequency, or at irregular intervals. Within agent computer 22, the sub-allocations typically are performed at regular intervals, though it will often be desirable to re-calculate socket allocations at a greater frequency than the applications allocations. As an example, agent allocations may be updated every 2 seconds, application allocations every half second, and socket allocations every 100 milliseconds.

The exemplary equations above apply primarily during steady state operation, when all involved sockets are open, have non-zero allocations, and are at least somewhat active (e.g., consuming some bandwidth). Typically, some provision or modification is made for startup conditions, and for conditions where sockets are idle and/or have socket allocations that are negligible or zero. For example, on occasion an agent computer may have launched applications that are not communicating with the network. In this case, the effective application priorities would all be zero because there is no socket activity, leading to an undefined result in equation (4). A straight average may be used to address this condition, in which case available bandwidth is apportioned among the applications equally. Equal apportioning may also be used in the case of undefined results in equations (2) and (6).

Referring now to equation (1), the APP-BWMGRs and/or other bandwidth managers may be configured to over-provision bandwidth. This may be done to ensure that truly idle sockets maintain an effective socket priority of zero, and thus do not obtain positive allocations (e.g., through applying the weighted average of equation (2)). Over-provisioning may also be used to provide internal allocations to idle sockets that are becoming active, so as to allow them to achieve a non-zero effective socket priority and thereby obtain a share of the application allocation being sub-allocated by the overlying APP-BWMGR 308.

Regardless of the particular methods employed, bandwidth may be apportioned among entities or components at a particular level based on socket status of a given component, or of components underlying that component. Socket status may include the number of sockets open within the domain of a given control module, on a given computing device, or for a given application. Equations (3) and (4) above provide an example of apportioning bandwidth among applications based partly on the number of sockets open at each application. All other things being equal, in these equations an application with more sockets open will have a higher effective priority and will receive a greater share of the allocated bandwidth. From the above, it will be appreciated that socket status may also include assigned priorities of sockets. As discussed at length above, assigned priority may be derived from a nearly limitless array of parameters, and may be set via network policies. The above equations provide several examples of allocation and sub-allocation being performed based on socket priority. All other things being equal, a socket with a higher socket priority, or an application or computer with higher priority sockets, will receive larger allocations in several of the above examples.

Socket status may also include consumption of past allocations, as should be appreciated from the above exemplary equations. The above examples provide numerous examples of shifting bandwidth allocations away from sockets, applications and computers consuming a lower portion of past allocations, relative to counteφart sockets, applications and computers.

Implementation of the above exemplary systems and methods may enable bandwidth resources to be allocated to high priority, critical tasks. Bandwidth may be shifted away from low priority uses, and/or consuming entities that are relatively idle in comparison to their counteφarts. The different embodiments above may be implemented to provide a more granular, or finer, level of control over bandwidth usage of a particular component within the system, hi particular, the bandwidth manager scheme described above may be configured to control socket and/or transaction level allocation for multiple conversations carried out by an application. For example, a given application may carry out a number of functions having varying levels of priority. The increased granularity described above allows system operators to assign relative bandwidth priorities to the various tasks carried out by a given application.

Finer control over bandwidth usage may be obtained by providing for dynamic modification of socket priorities after they have been assigned. As discussed above, socket priorities (e.g., socket priorities SP in equations (1) and (3) above) typically are assigned as sockets are opened on the associated computer, and the assigned priority can be derived from virtually any variety of policy parameters. Though the socket priorities may remain static while the socket is open, the agent module may be alternately configured to dynamically vary the socket priority when certain mission critical network tasks are being performed by the socket. These mission critical tasks may be defined through the policy mechanisms discussed above, using any desirable combination of policy parameters. Upon detection of a predefined task, the assigned priority for the socket is modified (e.g., by overriding the assigned value with a higher priority). This modification typically is employed to ensure that a desired level of service or resources is available for the predefined task (e.g., a desired allocation of bandwidth). Alternatively, the predefined task may be a low priority task, such that the socket priority may be dynamically downgraded, to ensure that bandwidth is available for more important tasks. Any of the above embodiments or method implementations may be adapted to provide for such dynamic modification of priorities.

The granular control at the transaction level (e.g., by the agent module embodiments discussed herein) enables the system to detect the commencement of critical transactions. Upon detection of such a transaction, the system is able to communicate through a chain of upstream components to ensure that a sufficient amount of bandwidth is reserved to complete the transaction within an acceptable amount of time. For example, a dynamic increase of socket priority typically will lead to increased allocations for not only the socket, but also for the application associated with the socket, and for the computing device on which the application is running (e.g., through application of the exemplary equations discussed above). Alternatively, bandwidth may be dynamically decreased for low priority transactions, which may also be communicated upstream to affect allocations at the various tiers.

For example, the above exemplary systems may be configured so that mission-critical bandwidth priority is assigned to only a specified portion of the data requested from a given web server application. In this example, the network policy could specify that when a browser on a client computer attempts to access a particular web page, that socket priorities are overridden to ensure that critical data is provided at a high level of service. Other data (e.g., trivial graphical information) could be assigned lower priority. More particularly, in commonly employed protocols, loading a single web page may involve ^' several "get" commands issued by a client browser application to obtain all of the data for the web page. The agent modules described herein may be configured with a policy that specifies that the agent module is to monitor its associated computer to look for "get" commands seeking predefined high priority portions of data on the web page. Upon detection of such a get command, the socket level priority would be dynamically adjusted for that portion of the web page data.

Another example which illustrates the benefit of increased granularity is use of a multi-media application to deliver audio, video and other data over a wide-area network. Where network congestion is high, it typically will be desirable to assign a higher priority to the audio stream. The increased granularity of the described system allows this prioritization through application of an appropriate network policy, and ensures that the video stream does not adversely affect the audio delivery in a significant way.

The systems described herein may be further provided with a layered communication architecture to facilitate communication between components. Typical implementations involve use of a communication library to present a communication abstraction layer to various communicating components. For a given communicating component, the abstraction layer is architecturally configured to hide the details of the communications mechanism and/or the location of the other communicating components. This allows the components to be transparent to the underlying transport mechanism used. The transport mechanism can be TCP, UDP or any other transport protocol. The transport protocol can be replaced with a new protocol without affecting the components and their operations.

The components communicate with each other by "passing objects" at the component interface. The communication layer may convert these objects into binary data, serialize them to XML format, provide compression and/or encrypt the information to be transmitted over any media. These operations typically are performed so that they are transparent to the communicating components. When applicable, the communications layer also increases network efficiency through multiplexing of several communication streams onto a single connection.

While the present embodiments and method implementations have been particularly shown and described, those skilled in the art will understand that many variations may be made therein without departing from the spirit and scope defined in the following claims. The description should be understood to include all novel and non-obvious combinations of elements described herein, and claims may be presented in this or a later application to any novel and non-obvious combination of these elements. Where the claims recite "a" or "a first" element or the equivalent thereof, such claims should be understood to include incoφoration of one or more such elements, neither requiring nor excluding two or more such elements.

Claims

CLAIMS:

1. A system for dynamically allocating bandwidth on a network link interconnecting a plurality of computers, including a control computer and multiple agent computers, where each agent computer is capable of running one or more applications employing one or more sockets, the system comprising: a control module adapted to run on the control computer; and plural agent modules, each being adapted to run on one of the agent computers, where each agent module is adapted to monitor socket status on its associated agent computer and repeatedly communicate with the control module based on such socket status, where the control module is adapted to determine, for each agent computer, an agent allocation of the bandwidth on the network link based on communications received from the agent modules, such that the agent allocation for one agent computer depends on the socket status of such agent computer relative to the socket status of the other agent computers, and where each agent module includes multiple tiers of bandwidth management components adapted to sub-allocate the agent allocation for the associated agent computer.

2. The system of claim 1, where the control module is adapted to dynamically and repeatedly update the agent allocations based on socket statuses of the agent computers relative to each other.

3. The system of claim 2, where the control module is adapted to update agent allocations such that agent allocations are decreased for agent computers consuming lower percentages of past agent allocations.

4. The system of claim 1, where each agent module is adapted, in the event that its corresponding agent computer is running multiple applications, to sub-allocate its agent allocation into an application allocation for each of those applications.

5. The system of claim 4, where the agent module is adapted to dynamically and repeatedly update the application allocations based on socket statuses of the applications relative to each other.

6. The system of claim 4, where each application allocation depends on how many sockets are open for the corresponding application, relative to other applications which are to receive a portion of the agent allocation.

7. The system of claim 4, where each application allocation depends on assigned priorities for sockets of the corresponding application, relative to other applications which are to receive a portion of the agent allocation.

8. The system of claim 4, where each application allocation depends on consumption of prior allocations for the corresponding application, relative to other applications which are to receive a portion of the agent allocation.

9. The system of claim 4, where each application allocation depends on how many sockets are open for the corresponding application, assigned priorities for those sockets, and on consumption of prior allocations by those sockets, relative to other applications which are to receive a portion of the agent allocation.

10. The system of claim 4, where each agent module is adapted, in the event of an application with multiple sockets, to sub-allocate the application allocation for such application into a socket allocation for each of the sockets.

11. The system of claim 10, where the agent module is adapted to dynamically and repeatedly update the socket allocations, depending on consumption activity of the sockets relative to each other.

12. The system of claim 10, where each socket allocation depends on an assigned priority for the corresponding socket, relative to other sockets which are to receive a portion of the application allocation.

13. The system of claim 10, where each socket allocation depends on consumption of prior allocations by the corresponding socket, relative to other sockets which are to receive a portion of the application allocation.

14. The system of claim 10, where each socket allocation depends on an assigned priority for the corresponding socket, and on consumption of prior allocations by such socket, relative to other sockets which are to receive a portion of the application allocation.

15. The system of claim 1, where each agent module includes an agent bandwidth manager and is adapted to launch an application bandwidth manager for each of multiple applications running on the agent computer, and launch a socket bandwidth manager for each of multiple sockets associated with those applications.

16. The system of claim 15, where the agent bandwidth manager is adapted to receive the agent allocation for its associated agent computer and sub-allocate the agent allocation, such that each application bandwidth manager underlying the agent bandwidth manager receives an application allocation for its associated application; and where each application bandwidth manager is adapted to sub-allocate the application allocation of its associated application, such that each socket bandwidth manager underlying the application bandwidth manager receives a socket allocation for its corresponding socket.

17. The system of claim 16, where each socket bandwidth manager is configured to report consumption data upstream to the application bandwidth manager overlying the socket bandwidth manager.

18. The system of claim 17, where each application bandwidth manager is configured to sub-allocate its application allocation based on effective socket priorities of the sockets underlying the application bandwidth manager, where the effective socket priority for a particular socket is based on an assigned priority of the socket and on usage of past socket allocations by the socket.

19. The system of claim 16, where each application bandwidth manager is configured to report consumption data upstream to the agent bandwidth manager overlying the application bandwidth manager.

20. The system of claim 19, where each agent bandwidth manager is configured to sub-allocate its agent allocation based on effective socket priorities of the sockets underlying the agent bandwidth manager, where the effective socket priority for a particular socket is based on an assigned priority of the socket and on usage of past socket allocations by the socket.

21. The system of claim 16, where each agent module is configured to dynamically and periodically update application allocations and socket allocations based on interactions between the agent bandwidth manager and the application bandwidth managers underlying the agent bandwidth manager, and between the application bandwidth managers and the socket bandwidth managers underlying the application bandwidth managers.

22. The system of claim 21, where the socket allocations are updated more frequently than the application allocations.

23. The system of claim 1, where each agent module is adapted to: assign a socket priority to a socket opened on the agent computer associated with such agent module; provide a socket allocation of bandwidth to the socket based on the socket priority; and detect whether the socket is being used to perform a predefined network transaction and, after such detection, modify the socket priority for such socket and update the socket allocation to account for such modification.

24. An agent module configured to be loaded onto a computer to manage network bandwidth usage by such computer, the agent module comprising: an agent bandwidth manager; and an application bandwidth manager, the agent module being configured to launch one such application bandwidth manager for each of multiple applications running on the computer, where the agent bandwidth manager and application bandwidth managers are configured to dynamically interact so as to sub-allocate network bandwidth available to the computer into individualized application allocations for each of the applications.

25. The agent module of claim 24, where each application allocation depends on a socket status of the corresponding application, relative to other applications which are to receive a portion of the network bandwidth available to the computer.

26. The agent module of claim 25, where each application allocation depends on how many sockets are open for the corresponding application, relative to other applications which are to receive a portion of the network bandwidth available to the computer.

27. The agent module of claim 25, where each application allocation depends on assigned priorities for sockets of the corresponding application, relative to other applications which are to receive a portion of the network bandwidth available to the computer.

28. The agent module of claim 25, where each application allocation depends on consumption of prior allocations for the corresponding application, relative to other applications which are to receive a portion of the network bandwidth available to the computer.

29. The agent module of claim 25, where each application allocation depends on how many sockets are open for the corresponding application, assigned priorities for those sockets, and on consumption of prior allocations by those sockets, relative to other applications which are to receive a portion of the network bandwidth available to the computer.

30. The agent module of claim 24, further comprising a socket bandwidth manager, the agent module being configured to launch one such socket bandwidth manager for each of multiple sockets open on the computer.

31. The agent module of claim 30, where the agent module is configured so that, when one of the applications running on the computer has multiple associated sockets, the application bandwidth manager associated with such application dynamically interacts with the socket bandwidth managers of those sockets so as to sub-allocate the application allocation into a socket allocation for each of the sockets.

32. The agent module of claim 31 , where the application bandwidth manager and socket bandwidth managers are configured to dynamically and repeatedly update the socket allocations in response to changing conditions at the sockets.

33. The agent module of claim 24, where the agent bandwidth manager and application bandwidth managers are configured to dynamically and repeatedly update the individualized application allocations in response to changing conditions at the applications.

34. The agent module of claim 24, where the agent module is adapted to: assign a socket priority to a socket opened on the computer on which the agent module is loaded; provide a socket allocation of bandwidth to the socket based on the socket priority; and detect whether the socket is being used to perform a predefined network transaction and, after such detection, modify the socket priority for such socket and update the socket allocation to account for such modification.

35. An agent module configured to be loaded onto a computer to manage network bandwidth usage by such computer, the agent module comprising: an agent bandwidth manager; and a socket bandwidth manager, the agent module being configured to launch one such socket bandwidth manager for each of multiple sockets open on the computer, where the agent bandwidth manager and the socket bandwidth managers are configured to interact so as to divide network bandwidth available to the computer into individualized socket allocations for each of the sockets, where the agent bandwidth manager and socket bandwidth managers are configured to dynamically and repeatedly update the socket allocations, based on consumption activity of the sockets relative to each other.

36. An agent module configured to be loaded onto a computer to manage network bandwidth usage by such computer, the agent module comprising: an agent bandwidth manager; an application bandwidth manager, the agent module being configured to launch one such application bandwidth manager for each of multiple applications running on the computer, where the agent bandwidth manager and application bandwidth managers are configured to interact so as to divide network bandwidth available to the computer into individualized application allocations for each of the applications; and a socket bandwidth manager, the agent module being configured to launch one such socket bandwidth manager for each of multiple sockets open on the computer, where, for an application with multiple sockets, the corresponding application bandwidth manager and socket bandwidth managers are configured to interact so as to sub-allocate the application allocation into a socket allocation for each of the sockets.

37. A system for dynamically managing bandwidth consumption on a network link by a computer, comprising: a software-based agent module configured to run on the computer, the agent module being further configured to: assign a socket priority to each of a plurality of sockets as those sockets are opened on the computer; sub-allocate overall bandwidth usable by the computer into a socket allocation for each of the plurality of sockets, where the socket allocation for each of the plurality of sockets depends on the socket priority of such socket relative to the socket priorities of the other sockets; and detect when one of the sockets is being used to perform a predefined network transaction and, after such detection, modify the socket priority for such socket and update the socket allocations to account for such modification.

38. The system of claim 37, where the predefined network transaction pertains to only a portion of data being requested by the computer from a network resource, such that the socket priority is modified while the socket is being used to obtain such portion of data, and is unmodified while the socket is being used to obtain other portions of the data being requested from the network resource.

39. A system for dynamically managing bandwidth consumption on a network link by a computer, comprising: a software-based agent module configured to run on the computer, the agent module being further configured to: assign a socket priority to a socket opened on the computer; provide a socket allocation of network bandwidth to the socket based on the socket priority; and detect whether the socket is being used to perform a predefined network transaction and, after such detection, modify the socket priority for such socket and update the socket allocation to account for such modification.

40. The system of claim 39, where the socket priority is restored to an unmodified value after completion of the predefined network transaction.

' 41. A system for dynamically allocating bandwidth on a network link interconnecting a plurality of computers, including a control computer and multiple agent computers, where each agent computer is capable of running one or more applications employing one or more sockets, the system comprising: plural agent modules, each being adapted to run on one of the agent computers and monitor and repeatedly report a socket status of its associated agent computer; a control module adapted to run on the control computer, where the control module is adapted to dynamically allocate bandwidth on the network link among the agent computers based on relative socket statuses of the agent computers as reported by the agent modules, each agent module being further adapted to: assign a socket priority to each of a plurality of sockets as those sockets are opened on the agent computer associated with such agent module; detect when one of the sockets is being used to perform a predefined network transaction and modify the socket priority for such socket during performance of the predefined network transaction.

42. A method of dynamically managing bandwidth in a distributed network system with a network link interconnecting a plurality of computers, comprising: assigning a socket priority to a socket, where the socket is associated with an application running on one of the plurality of computers; providing a socket allocation to the socket, where the socket allocation defines network bandwidth usable by the socket; monitoring network transactions performed using the socket to identify whether the socket is being used to perform a predefined network transaction; and dynamically modifying the socket priority after detecting that the socket is being used to perform the predefined network transaction, and updating the socket allocation to account for such modification of the socket priority.

43. The method of claim 42, further comprising: providing an application allocation to the application, where the application allocation defines network bandwidth usable by the application and is dependent upon the socket allocation; and dynamically modifying the application allocation to account for modification of the socket priority.

44. The method of claim 42, further comprising: providing a computer allocation to the computer on which the application is running, where the computer allocation defines network bandwidth usable by such computer and is dependent upon the socket allocation; and dynamically modifying the computer allocation to account for modification of the socket priority.