Classifications

• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F15/00—Digital computers in general; Data processing equipment in general
  • G06F15/16—Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
  • H04L67/00—Network arrangements or protocols for supporting network services or applications
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
  • H04L61/00—Network arrangements, protocols or services for addressing or naming

Definitions

• the present invention relates to network communications between n-to-n points in a network, where n is any integer value.
• the client-server paradigm is popular in the industry due to its simplicity in which a client makes a request and server responds with an answer.
• a popular communications protocol used between a client and a server in a communication network is, transmission control protocol/Internet Protocol, or simply, "TCP/IP.”
• TCP/IP transmission control protocol/Internet Protocol
• a client or client system or machine
• server or server system or machine
• a single physical server is often incapable of effectively servicing large number of clients. Further, a failed server leaves clients inoperable.
• cluster configurations having many servers running in parallel or grid to serve clients were developed using load- balancers.
• a conventional network level-clustering platform must be generic and usable by a wide range of applications. These applications range from, web-servers, storage servers, database servers, scientific and application grid computing. These conventional network level clusters must enable aggregation of compute power and capacity of nodes, such that applications scale seamlessly. Existing applications must be able to be run with minimal no o ⁇ changes. However, conventional network level clusters have had only limited success.
• SMP Symmetric Multi-Processor
• SMP Symmetric Multi-Processor
• TCP/IP The TCP's data delivery guarantee, ordered delivery guarantee and ubiquity, makes it particularly desirable for visualization.
• TCP's support for just two-end points per connection has limited its potential.
• Asymmetrical organization of processing elements/nodes that have pre-assigned tasks such as distributing incoming requests to cluster are inherently inflexible and difficult to manage and balance load.
• Asymmetrical nodes are often single point of failures and bottlenecks.
• MC Multi Computing
• Another problem with asymmetry in a client-server environment is latency.
• Switches and routers employ specialized hardware to reduce latency of data passing through.
• node's UDP/TCP/IP stack When data must pass through node's UDP/TCP/IP stack, it adds significant latency due to copying and processing.
• systems in order to achieve optimal performance, systems must avoid passing of data through intervening nodes having asymmetric organization.
• a server node's CPUs if a server node's CPUs must handle large amount of network traffic, application throughput and processing suffers.
• conventional systems must use hardware accelerators such as specialized adaptor cards or Integrated Circuit chips to reduce latency at the endpoints and improve application performance. This increases system costs and complexity.
• Low-cost fault-tolerance is a is highly desired by many enterprise applications.
• Solutions where fixed number of redundant hardware components are used suffer from lack of flexibility, lack of ability to repair easily and higher cost due to complexity. Solutions today offer high availability by quickly switching services to a stand-by server after fault occurred. As the stand-by systems are passive its resources only not utilized resulting in higher cost. In the simplest yet powerful form of fault tolerance by replication, the service over a connection continue without disruption upon failure of nodes.
• an active node performs tasks and passive nodes later update with changes. In many instances, there are fewer updates compared to other tasks such as query. Machines are best utilized when load is shared among all replicas while updates are reflected on replicas. Replica updates must be synclironous and must be made in the same order for consistency. With atomic delivery, data is guaranteed delivered to all target endpoints before client is sent with a TCP ACK indicating the data receipt. In the event of a replica failure, remainder of the replicas can continue service avoiding connection disruption to effect fault-tolerance. Non atomic replication lacks usability. Specifically, when a client request is received by replicas of a services, each produce a response.
• one conventional solution is a device for balancing load in a cluster of Web-Servers is popular in the industry.
• This load-balancing device which is also disclosed in U.S. Patent Numbers 6,006,264 and 6,449,647, switches incoming client TCP connections to a server in a pool of servers.
• a conventional server for this process is Microsoft's Network Load balancer software, which broadcasts or multicasts client packets to all nodes by a switch or router.
• the same server handles all client requests for the life of TCP connection in a conventional one-to-one relationship.
• a problem with conventional systems such as the ones above is when a service is comprised of different types of tasks running on nodes, it fails to provide a complete solution because any mapped server that would not run all services client would request over a connection results in service failure. This limits the use of such systems to web-page serving in which only one task of serving pages is replicated to many nodes.
• any mapping of devices implemented external to a server is a bottleneck and results in a single point of failure.
• replication is not supported. Therefore, with such single ended TCP, updates are not reflected on replicas, and hence, there are considerable limits on usability.
• TCP's current scope of host-to-host communication to group-to-group communication, more specifically extending current definition of two connection endpoints, to two groups of endpoints spanning symmetrically organized nodes. Each such endpoint is entitled to receive and transmit independently and in parallel while maintaining TCP's ordered transmission. Data is delivered to whole group or a subset depending on the configuration. Only necessary target end-points are required to have received the data, before TCP's ACK is sent to peer group.
• the present invention allows for addressing a cluster of nodes as a single virtual entity with one or more IP addresses.
• the communication between the client group and the server group is strictly standards based in that any standard TCP/IP endpoint is able to seamlessly communicate with the group.
• the data is delivered atomically to the endpoints tenninating at symmetrically organized nodes of the group.
• Filters installed at the connection endpoints filter out arriving data uninteresting to application segments. Data delivery to the application segments are dynamically controlled by the filters configured and installed appropriately. Additionally the filter optionally performs placement of incoming data directly into target application memory without intervening copies.
• nodes may receive and transmit independent of each other and in parallel. All transmissions are sequential per TCP specification, and transmission control among nodes are ordered based on round-robin among group nodes or round-robin among transmit requestors, or application specific schemes. Nodes may re-transmit in parallel and require no additional synchronizations between them to do so.
• application functions are distributed among group nodes. To achieve this, an application is logically segmented, each running a subset of the application functions. Incoming requests arriving on a TCP connection are then delivered to the segments that reside over the group effectively distributing load. By delivering only certain set of requests to application instances, a logical segmentation may be achieved without application code change. Application segments may be migrated from node to node dynamically without connection disruption.
• a node may communicate with other nodes of a group by creating a connection to the virtual entity represented by the group. This provides all the above features for communication between group nodes.
• the system is fault-tolerant in that if nodes nning an application fail, a set of remaining application replicas in the group continue service without disruption of the connections and service. Nodes may be added to or retired from the group dynamically, to maintain a certain quality of service in a manner transparent to the applications. For the purpose of balancing load among nodes or retiring a node, system transparently migrates active services and re-distribute tasks within the group.
• the system shares the processing load among a group of nodes by dynamically and transparently distributing incoming tasks over a connection to various application segments.
• a single request arriving over a connection may be serviced by multiple segments working in cohesion, enabling finer distribution of computation or processing among the nodes.
• the system allows for multiple instances of a segment run in parallel. Requests are delivered to the instances selected based on schemes such as round-robin, least loaded node, affinity based, content hashing.
• Incoming requests over a connection are delivered atomically to multiple segment instances for fault-tolerance. The results are optionally compared and a single instance is output. Upon failure of segments/nodes, remaining segment instances continue service without disruption of the connection.
• the system allows for flexible and external management of the system, by distributing tasks in a fine-grained fashion controlling and configuring filters at the connection endpoints.
• load responsibilities of the node are migrated to another node selected using schemes such as lowest loaded, round robin or an application specific scheme.
• the system automatically and dynamically adds resources to the group from a pool to meet changing needs.
• nodes are retired and provisioned dynamically and automatically.
• the system maintains specific quality of service adding or retiring resources automatically and dynamically.
• FIG. la is a generalized diagram of communication system constructed in accordance with one embodiment the present invention.
• FIG. lb is a block diagram illustrating a communication system in accordance with one embodiment of the present invention.
• FIG. lc illustrates a block diagram of organization of higher-level components for implementation of a communication system in accordance with one embodiment of the present invention.
• FIG. Id illustrates a block diagram of implementation of low-level organization components for optimized performance of a communication system in accordance with one embodiment of the present invention.
• FIG. 2 illustrates a block diagram of hardware organization of higher-level components for implementation of a communication system in accordance with one embodiment of the present invention.
• FIG. 3 a illustrates a flow chart for input data processing path on a connection in accordance with one embodiment of the present invention.
• FIG. 3b illustrates remainder of FIG. 3a, a flow chart for input data processing path on a connection in accordance with one embodiment of the present invention.
• FIG. 3 c illustrates a flow chart for filtering incoming data on a connection in accordance with one embodiment of the present invention.
• FIG. 4 illustrates a flow chart for transmitting data over a connection while limiting maximum transmission size at a time for fairness among nodes in accordance with one embodiment of the present invention.
• FIG. 5a illustrates a block diagram of a request/grant scheme for active sendHead assignments in accordance with one embodiment of the present invention.
• FIG. 5b illustrates a flow chart for processing a request for sendHead in accordance with one embodiment of the present invention.
• FIG. 6 illustrates a block diagram describing a virtual window scheme for peer group window advertisements in accordance with one embodiment of the present invention.
• FIG. 7a illustrates a block diagram of a computing system for a communication system in accordance with one embodiment of the present invention.
• FIG. 7b illustrates a block diagram of a computing system for a communication system having providing offloading of a main processor in accordance with one embodiment of the present invention.
• FIG. 7c illustrates a block diagram of a computing system for a communication system providing offloading of a main processor to dedicated hardware/accelerator chips in accordance with one embodiment of the present invention.
• FIG. 8 illustrates an alternate and generalized diagram of communication system in accordance with one embodiment of the present invention.
• FIG. 9 illustrates a data delivery and acknowledgement scheme between a client group and a server group in accordance with one embodiment of the present invention.
• FIG. 10 illustrates a logical view of an implementation in accordance with one embodiment of the present invention.
• FIG. 11 is a generalized diagram of a symmetric multi-computer system with fault tolerance, load distribution, load sharing and single system image in accordance with one embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION
• the present invention includes a system enabled for reliable and ordered data communication between two sets of nodes with atomic multi-point delivery and multi-point transmission, for example, extending TCP/IP.
• the present invention extends TCP's notion of reliable host-to-host communication to include symmetric group-to-group communication maintaining TCP specifications for data traffic between the groups. Further, the present invention extends the definition of two endpoints of a TCP connection to now include at least two groups of endpoints that are communicating over the single connection.
• end-points of a connection terminate at group nodes.
• each endpoint is comprised of a receiveHead and a sendHead operating independently.
• a node includes a connection on a network, for example, a data processing device such as general purpose computers or other device having a microprocessor or software configured to function with a data processing device for network related communications.
• a group refers to a collection of one or more nodes organized symmetrically.
• An application segment refers to the application or a segment of an application that may serve in conjunction with other application segments running on various group nodes.
• An application is comprised of one or more application segments and an application segment is comprised of one or more processes.
• a sendHead refers to a transmitting end of a TCP comiection, which controls data transmission and maintains the transmission state at the node.
• a receiveHead refers to the receiving end of a TCP connection, which controls data reception on connection and maintains data reception state at the node.
• An active sendHead refers to the sendHead that is designated to have latest transmission state information, for example, sequence number of data and sequence number of last acknowledgement.
• a bus controller refers to a node that controls and/or coordinates connection establishment and termination process with the peer group.
• a signal refers to a message exchanged within a node group over a logical bus.
• An end-point of a connection refers to stack such as TCP, that exchanges data with the peer in sequential order based on a pair of sequence numbers agreed beforehand.
• An end- point of a connection at least has an output data stream origination point and input data stream termination point.
• a request refers to a select segment of incoming data stream, for example, a client request for service.
• the communication system includes a TCP connection 130 that couples between a first group 120 and a second group 160.
• the first group 120 has a first, second, and third member nodes 100a, 100b, 100c and the second group 160 has a first and second member nodes 150x and 150y.
• the member nodes in either group are organized symmetrically in that each node has equal access to a TCP connection and operates independently and in parallel.
• a first data stream 110 and a second data stream 111 can flow between the first group 120 and the second group 160 of the communication system.
• a first application segment 135 and a second application segment 136 constitute a server application on 120.
• the first application segment 135 has a set of replicas 135x, 135y and the second application segment 136 also has a set of replicas 136x, 136y.
• the application segment replicas 135x and 135y rims over nodes 100a and 100b respectively while the replicas 136y and 136x runs over nodes 100b, 100c respectively.
• a client application at group 160 is comprised of an application segment 151 with replicas 151a and 151b.
• Application segments 135 and 136 of the first group 120 communicate over the connection 130 with segment 151 of the second group 160.
• the two data streams HO and 111 of the connection 130 follow TCP protocols.
• the connection 130 may have three different connection end points 130a, 130b, 130c at the first group 120 and two different connections end points 13 Ox and 130y at the group 160 on the same connection.
• Each group 120, 160 is assigned a respective group Internet Protocol ("IP") address 121, 161. Groups view each other as a single entity while being composed of nodes. Communications between two groups 120, 160 are addressed to each other through the group IP addresses 121, 161. When a request from say segment 151 arrives at the first group 120, it is viewed as data coming from group IP address 161. Similarly, the second group 160 sends data targeted to group address 121.
• IP Internet Protocol
• the endpoints 130a, 130b and 130c at the first group 120 may be set such that one or more of the application segment replicas 135a, 135b, 136a, 135b are delivered with an incoming request.
• Either of the endpoints 130x or 130y at the second group 160 may send request to server group 120.
• One or more of the receiveHeads at the endpoints 130a, 130b, 130c at the first group 120 receives the data depending on the settings.
• the endpoints 130a, 130b, 130c at the first group 120 may send response data which is received at the endpoints 130x, 130y at the second group 160.
• Application processes wanting to receive certain or all in coming data are guaranteed to have received it before acknowledging client with the receipt of data.
• the TCP sequence numbers are assigned in sequential order before data transmission starts.
• the communication system in accordance with the present invention provides client/server requests and responses that are beneficially atomic. That is, they are sent or received as a contiguous sequence of bytes, enabling multiple processes over two groups send and receive data over a single connection.
• the protocol between groups 120 and 160 is TCP and data is guaranteed to be delivered in the sequential order it was sent as per conventional TCP.
• TCP ACK segment indicating the receipt of data.
• replica outputs must be reduced to transmission of a single copy output, it is guaranteed that output is atomic in that data is transmitted if all nodes output same data.
• results don't match applications may optionally choose output to transmit based majority agreement or correct or successful result, etc.
• the first group 120 is comprised of the first, second, and third nodes 100a, 100b, 100c.
• the connection 130 between the first group 120 and the second group 160 has the outgoing and the incoming data streams 110, 111.
• Each node 1 OOa-c has a group-to-group communication stack 130a-c respectively.
• the delivery of data to all of the nodes is through a switch 141a-c coupled with the respective nodes lOOa-c. No assumption about the delivery guarantees to switch 141a-c by the underlying hardware is made, since popular hardware technologies such as Ethernet are unreliable. Delivery of data to each node lOOa-c or any of its subsets may be selective or no delivery at all is possible by the underlying hardware devices.
• the incoming data is switched by the switch 141a-c to either regular TCP/IP stack 140a-c or to the group-to-group communication stack 130a-c, based on the IP address and/or port.
• An application process 142 of node 100 communicates using the standard TCP stack 140.
• the application segments 135x,y, 136x,y communicate with group communication stack 130a-c respectively.
• the 105 carry control signals that coordinate and controls operations of group 131.
• the scope of the signals sent over control bus 105 is limited to the first group 120.
• the virtual bus 143 is comprised of the first and the second data streams 110, 111 and control signals 105 spanning group 120. This bus is directly tapped into by the peer group TCP connection 130.
• An alternative to the virtual bus 143 is the point to point communication between nodes and has the advantage of better bandwidth usage. However, this necessitates each node in a communication system to keep track of other nodes and their addresses and their roles.
• a logical bus model is preferred over control messaging due to location and identity transparency.
• connection end point 130a in accordance with one embodiment of the present invention.
• the switch 141 directs data to either standard TCP stack or the group-group communication stacks Internet Protocol ("IP") input 171.
• IP Internet Protocol
• IP Internet Protocol
• input packet is not fragmented, it may be passed directly to the input content filter 171 after few basic consistency checks.
• the input content filter 171 examines the input data content and or packet header to detennine if it contains data to be passed in to the application segment (e.g., 135x, 135y, or 136x).
• the communication system determines not to pass a packet further up, it is discarded with no further action and any memory is freed. Otherwise the input content filter 171 marks segments of the packet that is being passed into application. The packet is then passed to IP input processing layer 172 for complete validation including checksum computation and other consistency checks. Any invalid packets are discarded with no further processing. Resulting packets are then passed into a group-group TCP layer 173.
• the group- group TCP layer 173 coordinates with group nodes (e.g., 120, 160) and controls data receipt to meet TCP specification requirements such as acknowledgements to peer group.
• the group- group TCP layer 173 maintains the input TCP states of connection and passes data to socket through data path 137. Data path 138 indicates transmit data path from socket interfaces into the stack.
• the user socket sends out data invoking an output content filter 174.
• the output content filter 174 is not installed, and hence, performs no operation.
• a filter for fault tolerance synchronously compare data to be sent with other replica segment outputs and transmits a single output instance.
• the selection of output instance transmitted to peer group depends on the policy set in the filter such as equal outputs, majority agreement, correct result or successful operation output and the like.
• a replica Upon failure of a transmitting segment instance, a replica takes over and continues transmissions without connection disruption. At successful output instance reception at peer group, the replicas discard the data and frees up memory.
• the output content filter 174 passes data for transmission, to a group TCP output layer 175.
• the group TCP output layer 175 controls data transmission and maintain transmission states in conjunction with group nodes.
• the group TCP output layer 175 works with its group nodes to transmit data to peer group in the sequential order as specified by TCP.
• the group TCP output layer 175 passes an IP output layer 176 with data to transmit, liie IP output layer 176 later performs standard IP functions on the data and passes it down to device driver 177 for data transmission.
• an output comparison result by the output content filter 174 indicates differing outputs produced by nodes, a subset replicas are considered faulty and excluded from further service over connection while remaining endpoints continue service without connection dis ption.
• exclusion is based on schemes where majority of endpoints agree on a result to exclude others.
• exclusion of endpoints may occur where an operation failed.
• Exclusion of an endpoint may also be from any application specific scheme that is programmable with filter.
• connection end point 130 in a group-group communication stack where the content processor examine input and output data, in accordance with one embodiment of the present invention.
• Content filtration is a function of content processor. Content processors determine where in the application memory the data must be placed, order of data and time to notify application such as a full request is received. Working in conjunction with the network interface device driver 177, data is copied between a network interface 193 and an application memory 190 by a direct memory access controller 196. [0077] Examining incoming new request data, the content processor allocates the memory 192 in the application space. The allocation size is application specific, typically size of the complete request data from peer.
• Output data 193 is allocated by application itself. Further, there may be copies of segment of request/response data 194, 195. With this scheme application data is directly copied between network interface input/output buffer and application memory with no intervening memory copies involved.
• the first group 120 may be a set of servers comprised of the nodes 100 (100a,b,c).
• the second group 160 may comprise a set of client nodes 150 (150x, y).
• the nodes 100, 150 in each group 120, 160 are connected to each other via a connection device 180.
• the connection device 180 may comprise a broadcast/multicast device, for example, an Ethernet bus or a layer 2/3 switch.
• a network 189 may be any conventional network, for example, a local area network ("LAN”) or the Internet, through which two node groups are connected. The network 189 is not necessary when both peer groups are directly connected via the connection device 180.
• LAN local area network
• the 2 includes one or more network interface ports 185a,b,c at the server nodes 100a,b,c.
• Communication links 187a,b,c and 188a,b,c connect device 180 with nodes 100.
• the input data arriving through the connection end point 130 is replicated to 188a,b,c by the connection device 180 using its layer 2 or layer 3 multicast or broadcast capabilities.
• the arriving data is delivered to ports 185a, 185b, 185c. There are no guarantees of data delivery by 180, or the hardware ports or links involved.
• the data transmitted by the first group 120 through 187a, 187b, 187c are independent to each other, and hence, operate in parallel.
• the data transmitted through 187a, 187b, 187c to peer group are not necessarily visible to 120.
• signals sent over the logical bus 105 is replicated to links 188a, 188b, 188c by the device 180.
• Data sent to logical bus 105 of FIG. lb, is visible to server nodes 100a, 100b, 100c.
• Signals may have connection identification information common to a group. Further, signals may also and have source and target identifications. Target identification may be one or more nodes or may be an entire group.
• signals within the communication system may include an IACK signal, which is an input acknowledgement signal acknowledge input data from peer on behalf of the group.
• the IACK may include acknowledged sequence number, remaining bytes of data expected from peer group, window update sequence number, latest and greatest time stamp and a PUSH flag indicating if receiving active sendHead must send a peer group TCP ACK.
• a REQSH signal comprises a request and may ask for latest sendHead assignment targeted to an active sendHead.
• the target addresses may be an entire group.
• a GRANTSH signal comprises a message having active sendHead state information, bus time, list of nodes whose REQSH being acknowledged, and most recent IACK information known. A target of this signal assumes active sendHead after updating the state information.
• An IACKSEG signal comprises an input data acknowledgment, sent on behalf of a segment. It may have the information the same as or similar to the IACK signal.
• a REQJOLN signal is sent to application segments requesting to join the service over a connection.
• a LEAVE signal is sent requesting permission to leave service of an application segment on the connection.
• An ACKLEAVE signal grants permission to an application to leave servicing on a connection.
• a RESET signal is sent to request to reset a connection.
• a CLOSE signal is sent to request to close the output stream of connection by an application segment.
• An ACKCLOSE signal acknowledges receipt of CLOSE request. Connection establishment and termination
• TCP state diagrams are known. A flow chart describing such state diagrams is shown and described in a book by Richard Stevens entitled “TCP/IP Illustrated Volume I and Volume II,” the contents of which are hereby incorporated by reference. In addition, TCP/IP protocol and options are also discussed in RFC'793, and RFC 1323, the relevant portions of which are hereby incorporated by reference.
• a node is elected to act as the bus controller which co-ordinate and control the process and communicate with the peer group on behalf of the group.
• a static bus controller is chosen, however application program optionally selects bus controller as needed.
• bus contrqller function may be assigned to nodes in round robin fashion, alternatively bus controller may be chosen dynamically based on hashed value of in coming sequence number or source IP address/port address combination. A scheme where segment with lowest ID assume the bus controller role, when replicas of segments are available the bus controller responsibility is assigned on round-robin fashion among replicas.
• connection operation there are four types of connection operation in TCP. Each type follows different set of state transitions.
• active initiation When the group initiates a connection to peer group, it is referred to as active initiation, while a connation process in initiated by the peer group it is referred to as passive initiation.
• passive initiation Similarly when connection termination is initiated by the group, it is referred as active tennination and when termination is initiated by the peer group it is referred to as passive termination.
• the bus controller With passive initiation, upon arrival of a synchronization ("SYN") request from a peer group, the bus controller sends out REQJOLN signal requesting application segments to join the connection service. The bus controller then responds to peer group with an SYN request with an ACK (acknowledgement) for the SYN it received. When peer group acknowledges the SYN request sent on behalf of the group, the group nodes running application segments respond with a IACKSEG. When all required group nodes joined connection service with IACKSEG signal, connection is considered established and data transfer may be started.
• SYN synchronization
• the bus controller In active initiation, for a connection initiated from a group, the bus controller sends out REQJOLN signal targeted to group nodes. It then initiates connection process with peer group by sending SYN request on behalf of the group. Group nodes, upon receipt of a SYN request from peer group with an ACK for bus controller SYN, send IACKSEG indicating receipt of a valid ACK from peer group. Upon receipt of IACKSEG from required nodes, bus controller sends out ACK for the SYN request from peer group and the connection is considered established.
• the nodes With passive termination, upon receipt of FIN segment from the peer group, the nodes send a IACKSEG signal indicating the receipt of FIN.
• the bus controller responds to peer group with an ACK for the FIN (finished) received.
• the nodes finished sending data they send LEAVE signal indicating wish to leave connection.
• LEAVE request signals from all group nodes have been received after the FIN receipt, bus controller sends out FIN request to peer group.
• the bus controller sends out an ACKLEAVE signal and upon its receipt the target of the signal node enters the CLOSED state. Upon arrival of an ACK for the FIN request sent out, the bus controller enters CLOSED state.
• a data packet arrives on a node, it is checked (311) if the packet is targeted to a group address. If so and packet is a TCP fragment, fragment reassembly operation (314) is performed which yields a complete TCP segment upon the arrival of last fragment of the TCP segment. In most cases TCP segments are not fragmented so the no such operation is invoked.
• fragment reassembly operation 314 is performed which yields a complete TCP segment upon the arrival of last fragment of the TCP segment. In most cases TCP segments are not fragmented so the no such operation is invoked.
• standard TCP/IP stack is handed over (312) with the TCP segment for further processing.
• the receiveHead maintains state of input data such as if request is being ignored, passed in to application, start of a new request, need more data following to determine the target etc. As a packet may contain more than one request or partial request data, it is verified (330) if packet has remaining data to be processed. If there is no data left the filtration process is complete. [0097] When there is data remaining in packet to be filtered the current state is verified (331).
• request data If the current state indicates that request data must be discarded, up to a maximum of the request data in the packet is scheduled (332) as discarded and verified more any remaining data (330). Similarly if request data is being accepted and must be delivered to application segment, then remaining portion of the request data of the packet is scheduled for delivery. All delivered data must be check-sum computed, timestamp and packet header verified (333) only once and invalid packets are discarded (336) right away.
• the application specific filter is invoked (334) to determine data target and the current state updated to reflect the result of verification. If the filtration code could not determine the request target due to lack of sufficient data, it is combined with any immediately following data from reassembly queue which holds data arrived out of order. If there is still not sufficient data the remaining data is entered into reassembly queue so that the check is repeated when sufficient data arrives. Instated if sufficient data was found step 330 is repeated to filter data.
• Data input acknowledgement [0101] When atomic data delivery to multiple endpoints are required, acknowledgement for received data is sent out only when all endpoints have positively received the data.
• the target endpoints upon receipt of data sends out IACK signal over the bus indicating the data receipt in the TCP order.
• the active sendHead after verifying if all required nodes have received specific data, sends out TCP ACK segment to peer group if due per TCP specification.
• Multiple end points of a group may transmit data in TCP order. It is thus necessary to assign consecutive sequence numbers to segments of data to be transmitted. It is also necessary to maintain of the consistency of data transmitted, in order to avoid mixing up distinct request/responses from endpoints. For this purpose each complete request/response data is treated as a record by the transmitting node.
• the node to grant sendHead next is determined by selecting node with node-id that is numerically closest in a clock-wise manner from the list of REQSH requestors, highest priority sendHead requestor, round-robin or any application specific schemes. If however, more transmission records are awaiting assignment of sequence numbers step 387 is repeated with to send out remaining data.
• a node (100a) sends out REQSH (105r) signal requesting active sendHead role and active sendHead (100c) grants the role with GRANTSH (105t) signal with necessary state information to the requestor.
• the REQSH signal is sent out by Node 100a.
• Node 100b ignores the REQSH not being active an sendHead.
• Node 100c which is the active sendHead at the time of request, responds to 100a request with GRANTSH signal as sendHead is free to be granted.
• GRANTSH Upon receipt of GRANTSH signal, requesting node 100a assumes active sendHead.
• the GRANTSH signal contains a list of pending requestors as maintained by the group.
• Node 100b upon examining GRANTSH signal 105t, checks if its own request for active sendHead if any was acknowledged, by verifying list of pending requestors in the signal. When acknowledged, the retransmissions of REQSH signal is turned off.
• a node grants active sendHead to another, if it has outstanding data to be transmitted, it adds itself to the list of requestors to avoid additional request signals.
• An alternative to sending signals such as REQSH to all nodes is to send them directly to targets such as active sendHead node. The advantage of this approach is better bandwidth usage however, it lack location transparency.
• TCP Time stamp and round-trip-time (RTT) calculations in a group [0110] Most TCP implementations comply to The RFC 1323. It specifies a method to measure round trip time using time stamps. The round-trip time is typically measured subtracting the time-stamp of data server acknowledged from host's real time. To identify invalid packets due to wrapped around sequence numbers, the specification requires time stamps be monotonically increasing. [0111] Meeting the specifications with a number of nodes of varying types with different hardware timers is challenging. An ideal solution is nodes have a perfectly synchronized times, however is difficult at best. In one embodiment a specification requirement of monotonically increasing time stamp is met by synchronizing time on a node sending data with time of node sent data last.
• the basetime can be an arbitrary value chosen by bus controller initially and calculated thereon. Any time a node is granted with an active sendHead, the bustime of the grantor is sent along with the GRANTSH signal. The node granted with active sendHead, at the receipt of GRANSH signal, adjusts its bustime as set forth below.
• bustime grantor bustime (i.e. bustime from GRANTSH signal)
• bustime on nodes may not be perfectly synchronized with above formula due to GRANTSH transmission delay, it meets the requirement of monotonically increasing timestamp value.
• the granularity of bustime higher than the granularity of timestamp sent, error due to conflicting timestamps during concurrent retransmissions by nodes is reduced.
• timestamps have a granularity of 10 milliseconds and bustime having granularity of one microsecond, the error factor is reduced to one in ten thousand from one.
• the basetime at transmission is entered into transmission record by the sendHead.
• a fixed time value is added to the grantor bustime at the GRANTSH target node.
• TCP Window update in a group [0116] Window is the amount of data an end-point is able to accept data in to memory.
• Described here is a scheme where a group wide single virtual window size is used for effective window management in a group.
• the sendHead is responsible for updating peer group with window information on behalf of the group.
• Group nodes are initially assigned with the virtual window size.
• Nodes update window to active sendHead by sending input sequence number of data read-in by application once delivered.
• the active sendHead updates the peer group with the window, obtained by subtracting the amount outstanding data to be passed into application from the group wide virtual window size.
• Window updates are piggy-backed with IACK signal to reduce the number of window update signals. To further reduce the number of window update signals and TCP segments, a reserved window size is maintained in addition to the virtual window.
• the unacknowledged input sequence is indicated by 610, and 620 indicate the maximum data sequence number expected as advertised to peer group. 619 represent maximum reserved window sequence up to which a window update may be sent.
• 611, 612, 613, 614 shows the window sequences of data received by nodes 100a, 100c, 100b, 100c respectively.
• the 615 is the amount data that may be sent by the peer group.
• 617 and 618 shows the window updates sent by nodes 100a and 100c along with the IACK they sent with respect to 611, and 612.
• the maximum advertised window is shown by 621 and maximum reserved window is show by 622. Protection Against Wrapped-around Sequences with group TCP
• the TCP's current 32-bit sequence number wraps around in a short period of time.
• a delayed signal such as IACK, with wrapped around sequence number may be considered valid mistakenly when sequence number is used to acknowledge data input.
• IACK delayed signal
• the 64-bit sequence values used within the group are mapped back to 32-bit where it is used with peer.
• To map 32-bit sequence we split the 64-bit sequence into two 32-bit values where the least significant 32-bits represent the TCP sequence actively used with peer. The high order 32-bits count the number of times the sequence wrapped around since the connection was initiated. To map 64-bit value to 32-bit sequence number the least significant 32-bits used.
• IACK is sequenced although the overhead may be the same.
• Load among the replica segments may be shared by delivering requests to segments using schemes such as round-robin, least-loaded, hashed, affinity based etc. in conjunction with filters.
• Segment replicas enable fault-tolerance. If during input, should replicas fail, remaining replicas if any continue service without disruption. This is achieved by replicas maintaining a consistent view of the inputs.
• the segment controller enable the consistency among the replicas with atomic input delivery. A new segment controller may need be elected after failure.
• a replica fails during transmission of data, remaining replicas may continue the service without disruption of connection.
• Each replica agree on an instance of output and sendHead states are shared before transmission is started.
• Each replica frees up memory and data acknowledged by the peer group.
• Each application segment is free to choose number of replicas it maintains.
• a node dynamically elected as segment controller coordinates the replica IACK to form a segment IACK.
• the election of segment controller can be based on round-robin, least loaded, hashed or even a static node. Connection establishment, termination, window management all works as stated here in conjunction with corresponding schemes described earlier. In all cases when segment replicas agree on receipt of certain input segment, controller responds on behalf of the replicas. When the load is balanced among segment instances as opposed to replicas, no controller involvement may be necessary. [0126] When replicas receive data, they send IACK indicating receipt of input data.
• segment controller monitoring IACKs from replicas determines all replicas received certain input in the order sent, it sends out an IACK on behalf of the segment and initiates a client ACK.
• This IACK works as an acknowledgement to replicas that they pass the data to application socket or take any necessary actions atomically.
• Election of segment controller is round-robin per request or static to a select replica like the one with lowest replica ID. Node based group-to-group communication
• FIG. 7a it is a block diagram of a general computer and its elements suitable for implementing elements of invention.
• the group-to-group communication stack is executed by the processor(s) in the system.
• Group-to-group communication offloading the main CPU
• FIG. 7b it is a block diagram of a computer and its elements suitable for implementing elements of invention while offloading main processor from processing certain elements.
• the group-group communication stack is offloaded to an adaptor card with it own processor.
• FIG. 7c it is a block diagram of a computer and its elements suitable for implementing elements of invention while offloading main processor from processing certain elements of invention to dedicated hardware/accelerator integrated chips. The offloads most of the processing required otherwise by the main CPU by implementing the group-group communication stack fully or partially.
• a hardware device replicates a single TCP connection end point into multiple end points.
• the node group is represented by 802.
• the connection (801) has input stream 826 and output stream 825 respectively.
• the device (820) here is external to the nodes 800a,b,c.
• Each server nodes have connection end points 801a,b,c of the same connection 801.
• the device 820 replicates a single connection 801 into three (801a,b,c) end points on nodes 800.
• the device 820 has four ports 816, 817, 818, 819 of while port 819 is linked to the connection to peer device. This device is a potential single point of failure and adds extra network hop.
• FIG. 9 illustrated is an atomic data delivery and acknowledgement scheme between a client group and a server group where two data segments 910 and 911 must be delivered to two nodes 902 and 904.
• the 901 represent the client group and 902, 903, 904, 905 and 906 represent server group nodes.
• the 910 and 911 represent TCP the data segments sent by client group 901. Though 910 and 911 is potentially available at all server group nodes, however it is only delivered to nodes 902 and 904 as detennined by the input filtering system in this instance that may be programmed.
• Reference 912 represents the TCP ACK segment sent out to the client group from the server group sendHead 903.
• the plex controller 902 sends IACK signal 914 indicating atomic delivery only at the receipt of 913 PI ACK (Plex IACK) signal indicating that the acknowledgment of same data segments at the required plex/replica 904.
• the 902 does not send a PIAK since it is the controller responsible for sending IACK indicating atomic delivery of said data.
• the 903 having the sendHead upon receiving the IACK signal 914 sends out TCP ACK segment 912 to client group.
• ACK segment may optionally be sent out at the end of every complete request input. Also client ACK segment is sent out upon detecting exception conditions such as out of order segment arrival, timeout waiting for segment etc. so that client and server groups sync-up and retransmit quickly upon lost TCP segments. Should a server node fail to receive an IACK it sent PIACK for, it retransmits the PIACK and the active receiveHeads responds with another IACK indicating the latest sequence of input data it admitted in to node. [0132] Referring to FIG. 10, illustrated is a logical view of an implementation where input data is shared as in a bus however the output data is switched. The 1000 is the input data stream from the peer group.
• the 1010 is a logical half-bus where only input is shared using multicast or a shared media such as Ethernet.
• the 1020, 1021 and 1022 represent the bus input end-points to the nodes 1030, 1031 and 1032 respectively.
• 1040, 1041 and 1042 are the output end points that get fed into a layer 2 or layer 3 IP switching device 1050.
• the 1060 represent the aggregate output produced by the nodes 1030, 1031 and 1032 produced for input 1000.
• the 1000 and 1060 respectively forms input and output of a single connection.
• the server group (1112) is comprised of a number of nodes 1100 a,b,c,d,e,f.
• the input stream (1110) of TCP connection (1109) has multiple endpoints 1110a, b, c, d, e, f that span over the group nodes.
• the output stream (l l l l) of the same connection is comprised of endpoints 1111a, b, c, d, e, f.
• the application is comprised of three segments (1113, 1114, 1115) running over the entire group with two instances for each application segment 1113a,b, 1114a,b, 1115a,b.
• the segments are delivered with specific tasks based on criteria such as operations they perform, the data they manage.
• segmentation of application is achieved, in many cases without code change to existing applications.
• Applications may be segmented in many ways, examples include segmenting based on type or kind of requests a segment can handle, a hashing algorithm based on data content or connection information such as sequence number etc. It is also trivially possible that application is divided in to segments by programming them into different segments.
• the group nodes are paired as replicas 1100a,b, 1100c,d and 1100e,f such that each pair run two instances of the application segment 1113, 1114, 1115 respectively.
• Upon failure of a segment say 1100a the pair 1100b continue service without disruption. If failure of an instance say 1100a happen while transmitting, the other instance 1100b will send the remainder of the response to peer avoiding disruption of service.
• a new application segment instance may be added to a group so as to increase the fault-tolerance due added instance available to continue service in the face of failures. This may be done for example by creating a new process running application segment instance and then getting it added to group so that requests are distributed to it accordingly.
• non-empty subsets of groups are delivered with requests in specific orders such as round-robin and weighted priority that requests are essentially distributed among said non-empty subsets so as to balance the load on nodes.
• one or more replicas are delivered with a task, and after the task is complete the results from instances are sent out through the connection without regard for other replicas.
• one or more replicas may be delivered with same task. The relevant replicas then execute the operation in parallel and produces results.
• An output filter installed at the output stream of the group-to-group communication system compares results and a single instance of the result is sent out to peer group whereby the group appear as a single entity to peer group.
• the selection of output instance transmitted to peer group depends on the policy set in the filter such as equal outputs, majority agreement, correct result or successful operation output etc. Selection of the policy depends on the application.
• a replica takes over and continues transmissions without connection disruption.
• output comparison result by the output content filter indicates differing outputs produced by nodes
• a subset replicas are considered faulty and excluded from further service over connection while remaining endpoints continue service without connection disruption.
• exclusion of an endpoint such exclusion is based on schemes where majority of endpoints agree on a result to exclude others. Alternatively, exclusion of endpoints may occur where an operation failed. Exclusion of an endpoint may also be from any application specific scheme that is programmable with filter.
• the replicas are delivered with operations that result in state changes such as modified data in memory and storage. This way replicas maintain a consistent state.
• operations that does not affect consistency between replicas such as read operation, the task is delivered to only an instance of the replica. This enable balancing of load between the replicas. Node addition and retirement
• the filters at the connection end point of the TCP group-to-group communication system enable fine-grain control of data delivery to application segments.
• By dynamically configuring filters certain tasks are delivered to certain nodes, enabling external control over the delivery of task requests to node.
• flow of requests to application segments are controlled like a switch for fine task distribution among nodes.
• the group may be added with nodes any time.
• a newly added node may share load from existing connections and new connections.
• nodes join the service and starts accepting tasks arriving on it. When necessary load among nodes are balanced by migration of tasks.
• node retirement load responsibilities of the node are migrated to another, selected using schemes such as lowest loaded, round robin or an application specific scheme. While retiring, waiting for smaller tasks to finish while not accepting new tasks, the nodes are freed-up completely. When long running tasks are involved, the migration of tasks such as system level process migration is used. With process migration the entire context of application process such as stack, data open files are moved to another node transparently.
• Nodes communicate with other nodes of a group creating a connection to the address of the virtual entity represented by the group. This provides all the above features for communication between group nodes.
• the system automatically and dynamically adds resources to the group from a pool to meet changing needs. Similarly, nodes are retired and provisioned dynamically and automatically. The system monitors the quality of the service delivered to the clients and maintains specific quality of service adding or retiring resources. The operations can be done external to the system and are potentially transparent to the peer group.

Applications Claiming Priority (5)

Publications (2)

Family

ID=34397183

Family Applications (1)

Country Status (6)

Citations (2)

Family Cites Families (10)

• 2004
  • 2004-09-21 JP JP2006527145A patent/JP2007520093A/ja active Pending
  • 2004-09-21 EP EP04784749A patent/EP1668527A4/de not_active Withdrawn
  • 2004-09-21 WO PCT/US2004/031020 patent/WO2005031588A1/en active Application Filing
  • 2004-09-21 AU AU2004277204A patent/AU2004277204A1/en not_active Abandoned
  • 2004-09-21 KR KR1020067005585A patent/KR20060090810A/ko not_active Application Discontinuation
  • 2004-09-21 CA CA002538084A patent/CA2538084A1/en not_active Abandoned

Patent Citations (2)

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events