AU2005291837A1 - Peer signalling protocol and system for decentralized traffic management - Google Patents

Description

WO 2006/037163 PCT/AU2005/001514 1 PEER SIGNALLING PROTOCOL AND SYSTEM FOR DECENTRALIZED TRAFFIC MANAGEMENT The present invention relates to a method and a signalling protocol for decentralized traffic management in a computer network. 5 Small networks, comprising multiple endpoint devices, such as desktop computers, notebook computers and handheld computers, usually share a link to the external network, such as the Internet or Wide Area Network (WAN), via a common access device, such as a DSL modem and/or router. A common problem is that one endpoint can often occupy more than its fair 10 share of the available bandwidth. This is particularly true when one or more endpoint users are downloading large files, while another endpoint user is attempting to use delay-sensitive networked applications, such as voice over IP (VolP), video streaming or online games. It is known to solve these problems by deploying a so-called traffic 15 management appliance in series with the access device. These appliances are hardware devices that optimize network performance, through one or more of the following functions: traffic shaping, also known as bandwidth management or quality of service (QoS), compression, caching and protocol optimization. Alternatively, traffic management functions may be integrated into an existing 20 networking device, such as a router, obviating the need for a separate device. However, traffic management hardware devices are expensive. Routers that implement traffic management functions are also expensive. Network access devices found in small networks, however, are relatively inexpensive products lacking quality of service (QoS) capabilities. The cost of 25 installing the additional hardware or more sophisticated network access devices that can implement traffic management is often prohibitive and also introduces another point of failure into the network.

WO 2006/037163 PCT/AU2005/001514 2 It is an object of the present invention to provide a method and a signalling protocol for traffic management that overcomes or at least substantially ameliorates the problems associated with the prior art. Other objects and advantages of the present invention will become apparent 5 from the following description, taken in connection with the accompanying drawings, wherein, by way of illustration and example, embodiments of the present invention are disclosed. In one form of this invention there is proposed a network traffic management system for managing traffic flows to a network resource including 10 a plurality of endpoints adapted to send traffic to the network resource, means to set and communicate a policy for access to the network resource by said endpoints, a software agent associated with each of said endpoints adapted to communicate with each other said agent, and to negotiate access to the s15 network resource for the respective endpoint in accordance with the policy and the information communicated by each other agent. In a further form the invention may be said to reside in a method of shaping traffic presented to a network resource in a local area network including the steps of 20 providing software agents at each of a plurality of network endpoints on a local area network, said agents being adapted to communicate with each other agent by signals passed over the local area network; providing at each endpoint means to regulate the traffic presented to the network resource by that endpoint; 25 providing and communicating to each agent a policy for the allocation of use of the network resource among the plurality of endpoints; each agent communicating with each other agent to negotiate access to the network resource according to the policy and the communicated traffic requirements of each endpoint.

WO 2006/037163 PCT/AU2005/001514 3 In a yet further form, the invention may be said to reside in a network communications protocol adapted to allow endpoints in a local area network to communicate their respective requirements for access to a shared network resource to each other. 5 In preference, the protocol includes means for endpoints to identify each others presence on the network. In preference, the protocol includes means to communicate relative priority information between endpoints. In preference, the protocol includes means to communicate a policy for access 10 to the shared network resource among all endpoints. In preference, there resides on the local area network at least one endpoint adapted to set and communicate said policy for access. This endpoint is called a manager. In preference, the protocol includes means to communicate any or all of an 15 endpoint's requested bandwidth, the data rate of the network resource, network statistics or network data. In preference, the protocol includes means to implement a distributed cache amongst a plurality of endpoints. In preference, the network endpoints are computing devices. 20 In preference, the network endpoints are client devices such as desktop computers, laptop computers and handheld computers. In preference an endpoint is a computer server.

WO 2006/037163 PCT/AU2005/001514 4 In preference an endpoint is a computer peripheral, such as a printer or scanner. In preference the network resource is an access point to a wide area network. In preference, the network resource is an access point to the internet. 5 In preference, the policy for allocation of access to the network resource is a policy for allocation of a portion of bandwidth on a wide area network or internet connection to each endpoint, calculated as total available bandwidth available on the link divided by the number of endpoints. In preference, the policy for allocation of access to the network resource is a 10 policy for allocation of a portion of bandwidth on a wide area network or internet connection to each endpoint, based on pre-set or communicated relative priority of each endpoint. In preference, each endpoint includes means to implement the requirements of the policy by rate limiting. 15 In preference, the local area network may be a wireless network or a 3G network. For a better understanding of this invention it will now be described with respect to preferred embodiments which shall be described herein with the assistance of drawings wherein; 20 Figure 1 is diagram of a local area network upon which the method and a signalling protocol for decentralized traffic management of the invention is implemented; and Figure 2 is a diagram of the network of Figure 1, showing data flows when the invention is implemented; and WO 2006/037163 PCT/AU2005/001514 5 Figure 3 is a table showing a basic set of protocol signals for an embodiment of the Peer Signalling Protocol of the invention; and Figure 4 is a table setting out the header fields of the Protocol of the invention; and 5 Figure 5 illustrates time slicing for a synchronous embodiment of the invention. Now referring to the illustrations, and in particular to Figure 1, there is shown a desktop computer 10, a server computer 11 and a laptop computerl2. These are network endpoints, which share access to an external network 13 via a local area network (LAN)14 and a common access device, being a DSL modem o10 and router 15. In other embodiments (not shown) the endpoints may be any computing devices and the common access device may be a router, cable modem or network switch. The external network 13 shown as a corporate wide area network (WAN), may be the internet. The network endpoints on the Local Area Network (LAN) signal each other so 15 as negotiate access to the shared external network. The system thereby implements a virtual centralized traffic manager, by means of decentralized traffic management. This is possible because of the large disparity between LAN and WAN data rates, which is typically 2 or 3 orders of magnitude. LAN data rates are between 20 100Mbps andl0Gbps, whereas WAN data rates are typically under 45Mbps (T3) for a corporate WAN, or 1Mbps for a typical DSL Internet link. In the time it takes to transmit 1 Mbyte of data over a 1 Mbps WAN link, 100Mbytes of data can be transmitted over a 100Mbps LAN, which is equivalent to one million 100 byte messages. 25 The high data rates of the LAN make it feasible to implement a signalling protocol between endpoints that enables endpoints to negotiate access to the WAN and adjust on the fly what they are transmitting over the WAN.

WO 2006/037163 PCT/AU2005/001514 6 Essentially, the fast LAN is used as a signalling channel to control access to the much slower WAN. Signalling is the exchange of information between peer endpoints on a fast LAN to control access to a slower WAN. The large disparity in speed means that the bandwidth used by the signalling does not have a 5 detrimental effect on the data rate presented to the WAN. Figure 2 is a logical block diagram of the system of Figure 1. As illustrated in Figure 2, each endpoint includes software modules21 to implement a decentralized traffic management system. These software modules communicate by sending data signals 23 to each other across the 10 LAN 14. This data comprises signals 23 which implement the Peer Signalling Protocol (PSP) of the invention. Each endpoint so equipped 10, 11, 12 is referred to as a PSP host, and all PSP hosts are peers. There is a distributed system in which endpoints emulate and perform the functions of a centralized traffic management appliance. 15 Applications running on the endpoints communicate data 25 which is carried by the WAN and also data 26 which is carried only by the LAN. Endpoints signal their network traffic requirements to their peers, and each endpoint listens for the PSP signals from other peers. In order to send and receive data over the WAN, each endpoint must gain 20 access to the WAN bandwidth. Endpoints negotiate their use of WAN bandwidth according to a policy. In policy-based networking such as this, a policy is a formal set of statements that define how the network's resources are to be allocated among network users, hosts, applications and other devices. The policy may be static and pre 25 configured, or dynamic and communicated via the signalling protocol or another mechanism.

WO 2006/037163 PCT/AU2005/001514 7 One or more of the endpoints may be designated managers. These determine and broadcast the policy in effect to their peers. In the absence of a manager, there may be a default (pre-defined) policy, such as, "highest priority host gets 50% of the bandwidth". Note that while the use of 5 a manager centralizes the policy control, the policy enforcement is decentralized. Each endpoint controls the rate at which it send data to the WAN itself, in response to the set policy. This control is performed by a rate limiter 28. The major signals of this protocol are shown in Figure 3, but many more may 10 be contemplated. As shown in the figure the signals are as follows: transmit There are two defined time periods, TransmitPeriod and TransmitWait. All PSP hosts, while attempting to transmit traffic regularly broadcast a transmit messages every TransmitPeriod milliseconds, providing another host with 15 higher priority has not transmitted during the last TransmitPeriod milliseconds. A PSP host will refrain from broadcasting transmit messages if another host with higher priority is transmitting. A transmit message may not be broadcast less than TransmitWait milliseconds prior to the last message. The body of the transmit message may specify parameters, such as the 20 requested bandwidth, priority, and other information as required, or it may be empty. set The set operation is sent out by a PSP Manager to establish a policy. A PSP Manager is an endpoint configured to load and deploy policies. The body of the 25 set message contains the policy to be established.

WO 2006/037163 PCT/AU2005/001514 8 In the illustrated embodiment, the default policy is for a lower priority host to simply back off in the presence of traffic from a higher priority host. info The info operation indicates that an endpoint has new information to share, 5 such as network statistics or the WAN data rate. get This is sent as a request to get information, such as network statistics or cached data. pause o10 This is sent to request an endpoint or endpoints to pause transmissions. In the illustrated embodiment, PSP is a UDP-based broadcast protocol. The PSP header is shown in Figure 4. All integers are in standard network representation.

WO 2006/037163 PCT/AU2005/001514 9 Over Ethernet, the minimum PSP packet is 70 bytes (560 bits), broken down as follows: OSI Layer Header Size Layer 5 PSP header 30 bytes Layer 4 UDP header 8 bytes Layer 3 IP header 20 bytes Layer 2 Ethernet 12 bytes frame It can be seen, for example, that 100 hosts broadcasting10 times/second generate 560 kbps of LAN traffic, i.e., less than 1% of the capacity of a 5 100Mbps LAN. The transmission time for each signal is 5.6 microseconds. The usage of LAN resources does not therefore impact on the capacity of the LAN to present data to the WAN at the full speed which the WAN can accept. The system flexibly partitions the WAN bandwidth among the endpoints. The exact partitioning of bandwidth depends on the installed policy, and may be fair 10 or unfair, static or dynamic. The goal of the system is to ensure that the sum of the endpoint data rates is always equal to the overall WAN data rate, RWAN, as specified by the following rate equation. N Ri = RWAN where there are N endpoints, Ri is the rate of endpoint i and RWAN is the WAN 15 data rate. The system attempts to maintain the equality of the left and right sides of the rate equation in both send and receive directions at all times. Each WO 2006/037163 PCT/AU2005/001514 10 endpoint enforces its rate, Ri, by means of a bidirectional rate limiter 28 that controls the maximum rate of traffic sent and received. The rate limiter need only limit traffic 25 sent to or received from the WAN, not traffic 26 that is internal to the LAN. 5 If the sum of the endpoint rates is less than the actual WAN data rate that means that WAN bandwidth is under utilized. If the sum is greater than the actual WAN data rate that means that the WAN link is over utilized or congested. A congested WAN link means that endpoints are competing for bandwidth, rather than negotiating a guaranteed share of the bandwidth. 10 For example, a fair, static partitioning across N endpoints simply limits each endpoint to precisely 1/INth of the WAN data rate, i.e.: R, = R 2 = R 3 ... = RN = (RwAN/N) An unfair, static portioning allocates different rates to each endpoint, while nevertheless maintaining the equality of the rate equation. 15 Static partitioning can be implemented without any form of signalling but results in wasted bandwidth whenever one or more of the endpoints require less than their share. A dynamic partitioning scheme allocates bandwidth across the subset of endpoints that are actually transmitting at any given time. 20 Endpoints can determine what other endpoints are transmitting either implicitly by sniffing what is on the physical network, or explicitly via signalling. In the invention, endpoints broadcast or multicast PSP transmit signals periodically whenever they are transmitting. By counting the number of transmitters present during a given interval, the system can determine the total number of endpoints 25 that are currently transmitting and partition bandwidth accordingly.

WO 2006/037163 PCT/AU2005/001514 11 Further, by including a host priority in the transmit signal, lower-priority endpoints can be configured to back off transmitting either partially or completely in the presence of high-priority traffic. This can be used to implement a form of priority queuing. 5 A further embodiment includes an application priority in the transmit signal to enable priority queuing by software application, either in combination with host priority or independently. A further embodiment includes an endpoint's requested bandwidth in the transmit message. The requested bandwidth can be pre-configured or can be 10 computed dynamically by summing the minimum required bandwidth required for each networked application running on a given endpoint. The minimum required bandwidth for each application can be specified in the policy. Prior art techniques, such as windows-based traffic management, can be used to determine what networked applications are running on a given endpoint at a 15 given time. The system will then attempt to allocate each endpoint its requested bandwidth in order of priority. Each endpoint first sums the requested bandwidth from each transmit signal. If more bandwidth is requested than is available, endpoints will renegotiate their bandwidth requirements according to the installed policy and re-transmit accordingly. For example, one 20 policy could require endpoints to proportionally reduce their requested bandwidth, i.e., in a pro rata fashion. Another, policy could require low-priority endpoints to temporarily stop transmitting. A further embodiment specifies a minimum guaranteed bandwidth for each endpoint, but permits endpoints to burst above their minimum if not all of the 25 WAN link is being utilized. Endpoints signal their intent to use additional bandwidth, to avoid multiple endpoints from attempting to overallocate bandwidth.

WO 2006/037163 PCT/AU2005/001514 12 An embodiment with improved efficiency can be implemented by synchronizing the system time of each endpoint and assigning each endpoint a unique signalling time slot, as described in Figure 5. For example, if there are 100 endpoints transmitting 10 times/second (i.e., every 100 milliseconds) then 5 signals can be spread out every millisecond. Synchronization can be performed via a standard synchronization protocol, such as Network Time Protocol (NTP). Endpoints periodically determine the WAN data rate, i.e., the speed of the WAN link. The WAN data rate may be dynamic (vary over time). 10 In order to share the WAN bandwidth, it is first necessary for the system to know the WAN data rate (WAN link speed). Further, certain WAN technologies, such as Frame Relay, have variable data rates. In such cases the system must distinguish between the minimum guaranteed rate, sometimes called the Committed Information Rate (CIR), and the maximum access rate, sometimes 15 called the Excess Information Rate (EIR). When the WAN data rate is static and known ahead of time, endpoints can be simply pre-configured with this information via the installed policy. When the WAN data rate is dynamic or static but unknown, it must be computed. The system computes the data rate by performing a speed test. 20 Prior art describes speed testing for a single host. The method herein is for performing speed testing in a distributed system. The speed test involves transferring data to or from one or more known fast servers, herein known as speed test servers. (On the Internet, there are public speed test servers, such as DSLreports.com. It is also possible to set up and use a speed test server on 25 a corporate network.) The data rate is computed by clocking the data transfer and dividing the data size in bits by the transfer time in seconds minus connection/disconnection time WO 2006/037163 PCT/AU2005/001514 13 (if using TCP or another connection-oriented protocol. The key requirement is that there is sufficient data and both the sending and receiving hosts are sufficiently fast to completely saturate the WAN link. Several methods may be used to compute the WAN data rate in a peer 5 signalling system, two of which are described below. In the dedicated speed test, a designated endpoint, known as the tester, requests other endpoints to briefly pause transmitting, then performs a speed test. The tester first sends a pause signal to the other endpoints to request quiet time. Endpoints receiving a pause interpret may elect to delay or ignore 10 the pause request. Therefore the tester must wait until all other endpoints cease transmitting, or give up and try again later if transmissions continue unabated. Once the network is quiet, the tester performs a speed test, possibly in both directions. Upon computing the WAN data rate, which typically takes no more than a couple of seconds, the tester communicates the results via the 15 info or set signal. In a system with a designated manager, the manager listens for info signals and will set the new WAN data rate. In the piggyback speed test, the speed test is performed by clocking part of an existing data transfer. Any endpoint can perform the test providing the following 20 two conditions are met: 1) no other endpoints transmit during the speed test (determined by absence of other transmit signals) 2) the data transfer saturates the WAN data link Upon successfully completing the speed test, the results are communicated via 25 the info or set signal.

WO 2006/037163 PCT/AU2005/001514 14 For example, a piggyback speed test may be combined with downloading an operating system update or updating a local cache. There is a variety of prior-art techniques for determining when a WAN link is saturated. For example, one active measurement technique involves injecting 5 multiple Internet Control Message Protocol (ICMP) 'ping' packets into the WAN and measuring inter-packet delay, round trip times, and packet losses. The speed test server may also be utilized for performing active measurements. Endpoints that implement the peer signalling protocol are referred to as signalling devices. In a fully managed LAN, all endpoints should be configured 10 to be signalling, to ensure that the traffic contributed by all endpoints is taken into account. Traffic originating from signalling devices can be identified by signing packets, namely, adding a signature to each packet. One signing technique is packet colouring or packet marking, i.e., writing specific packet header fields, for example, the IP Layer 3 Type of Service (ToS) 15 or Differentiated Services (DiffServ) field. Marks may be altered dynamically to minimize the risk of spoofing by non-signalling devices. Another signing technique is encapsulation, which involves encapsulating the entire packet within another protocol, for example, by using Generic Routing Encapsulation (GRE). While the main use of GRE is to route IP packets 20 between private IP networks across an internet, it is also feasible to encapsulate packets over a LAN. Further, the encapsulation may include a digital signature for strong security. Conversely, non-signalling endpoints will not sign packets, and can therefore be identified by the absence of a packet signature, i.e., the correct packet mark or 25 encapsulation. Each endpoint in the system may be optionally configured to store data that is requested from a data source via the WAN, thereby creating a distributed WO 2006/037163 PCT/AU2005/001514 15 cache. Endpoints requesting data via the WAN can first request if data is available from one of their peers via a get signal. If a peer already has the data, the data can fetched quickly via the LAN, thereby eliminating redundant data transfers via the slower WAN. 5 Each endpoint in the system may optionally store numerous network statistics, such as bandwidth, round-trip time, latency, etc. Statistics for the system as a whole can then be aggregated to one or more designated reporter, specified by the policy. The reporter may be local to the LAN or accessible via the WAN. A possible aggregation method is Push-based aggregation wherein each 10 endpoint will periodically push statistics to the reporter via an info signal. Alternatively Query-based aggregation may be used wherein The reporter will periodically query each endpoint with a get signal and request the latest statistics. Each endpoint in the system may be optionally configured via the installed 15 policy to detect and respond to anomalous network conditions. For example, so-called port scanning or Denial of Service (DOS) style attacks, resulting in hosts creating and destroying large numbers of connections per second, are characteristic of network viruses. Such conditions can be detected and trigger self-policing behavior, for example, rate limiting some or all traffic originating 20 from the infected host. Although the invention has been herein shown and described in what is conceived to be the most practical and preferred embodiment, it is recognised that departures can be made within the scope of the invention, which is not to be limited to the details described herein but is to be accorded the full scope of 25 the appended claims so as to embrace any and all equivalent devices and apparatus.

Claims (21)

1. A network communications protocol for use in a computer network including a shared network resource and a plurality of endpoints adapted to send traffic to the network resource, 5 said protocol adapted to allow endpoints in the network to communicate their respective requirements for access to the network resource to each other; and to communicate a policy for access to the network resource by said endpoints, 10 and to allow negotiation of access to the network resource for the respective endpoint in accordance with the policy and the requirements information communicated by other endpoints.

2. The protocol of claim 1 wherein the protocol includes means for endpoints to identify each other's presence on the network. 15

3. The protocol as in any one of the preceding claims wherein the protocol includes means to communicate relative priority information between endpoints.

4. The protocol as in any one of the preceding claims wherein the protocol includes means to communicate any or all of an endpoint's requested 20 bandwidth, the data rate of the network resource, network statistics and network data.

5. The protocol as in any one of the preceding claims wherein the protocol includes means to implement a distributed cache amongst a plurality of endpoints. 25

6. The protocol as in any one of the preceding claims wherein the network resource is an access point to a wide area network. WO 2006/037163 PCT/AU2005/001514 17

7. The protocol as in any one of the preceding claims wherein the network resource is an access point to the internet.

8. The protocol as in any one of the preceding claims wherein the policy for allocation of access to the network resource is a policy for allocation of a 5 portion of bandwidth on a wide area network or internet connection to each endpoint, calculated as total available bandwidth available on the link divided by the number of endpoints.

9. The protocol as in any one of the preceding claims wherein the policy for allocation of access to the network resource is a policy for allocation of a 10 portion of bandwidth on a wide area network or internet connection to each endpoint, based on pre-set or communicated relative priority of each endpoint.

10. A network traffic management system for managing traffic flows to a network resource including 15 a plurality of endpoints adapted to send traffic to the network resource, means to set and communicate a policy for access to the network resource by said endpoints, a software agent associated with each of said endpoints adapted to communicate with each other said agent, and to negotiate access to the 20 network resource for the respective endpoint in accordance with the policy and the information communicated by each other agent.

11. The network traffic management system of claim 10 wherein the network endpoints are computing devices.

12. The network traffic management system of claim 10 or claim 11 wherein 25 the network endpoints are client devices such as desktop computers, laptop computers and handheld computers. WO 2006/037163 PCT/AU2005/001514 18

13. The network traffic management system of any one of claims 10 12wherein an endpoint is a computer server.

14. The network traffic management system of any one of claims 10 -13 wherein an endpoint is a computer peripheral, such as a printer or 5 scanner.

15. The network traffic management system of any one of claims 10 14wherein each endpoint includes means to implement the requirements of the policy by rate limiting.

16. The network traffic management system of any one of claims 10 -15 10 wherein each endpoint includes means to implement the requirements of the policy by rate limiting.

17. The network traffic management system of any one of claims 10 16wherein the local area network is a wireless network or a 3G mobile telephony network. 15

18.A method of shaping traffic presented to a network resource in a local area network including the steps of providing software agents at each of a plurality of network endpoints on a local area network, said agents being adapted to communicate with each other agent by signals passed over the local area network; 20 providing at each endpoint means to regulate the traffic presented to the network resource by that endpoint; providing and communicating to each agent a policy for the allocation of use of the network resource among the plurality of endpoints; each agent communicating with each other agent to negotiate access to 25 the network resource according to the policy and the communicated traffic requirements of each endpoint. WO 2006/037163 PCT/AU2005/001514 19

19. The method of claim 18 wherein there resides on the local area network at least one endpoint adapted to set and communicate said policy for access.

20. The method of any one of claims 18-19wherein each endpoint includes 5 means to implement the requirements of the policy by rate limiting.

21. The method of any one of claims 18-20 wherein the local area network is a wireless network or a 3G mobile telephony network.