Network adapter with TCP windowing support

• US 6,788,704 B1
• Filed: 08/05/1999
• Issued: 09/07/2004
• Est. Priority Date: 08/05/1999
• Status: Expired due to Term

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Lindsay, Steven B.
• Primary Examiner(s): Ton, Dang
• Assistant Examiner(s): HOM, SHICK C
• Application Number:
  US09/369,500
• Time in Patent Office: 1,860 Days
• Field of Search: 370/229, 370/230, 370/235, 370/395.4, 370/395.41, 370/395.42, 370/395.6, 370/395.7, 370/395.71, 370/465, 370/468, 709/102, 709/103, 709/104, 709/105, 709/106, 709/220, 709/221, 709/222, 710/260, 710/266, 710/267, 714/746, 714/748, 714/749, 714/750
• US Class Current: 370/465
• CPC Class Codes: H04L 1/1671 the supervisory signal being transmitted together with control information
• CPC Class Codes: H04L 1/1832 Details of sliding window management
• CPC Class Codes: H04L 1/1835 Buffer management
• CPC Class Codes: H04L 47/10 Flow control; Congestion control
• CPC Class Codes: H04L 47/193 at the transport layer, e.g. TCP related
• CPC Class Codes: H04L 47/27 Evaluation or update of window size, e.g. using information derived from acknowledged [ACK] packets
• CPC Class Codes: H04L 47/43 Assembling or disassembling of packets, e.g. segmentation and reassembly [SAR]
• CPC Class Codes: H04L 49/90 Buffering arrangements

Abstract

A network adapter and corresponding method for its use are disclosed. The network adapter has an operational mode that allows a host CPU to offload transmission of a block of data to the adapter. The adapter segments the block into fragments, and builds a data packet for each fragment. The adapter transmits these packets with an adapter-implemented flow control. This flow control uses: a context engine that tracks flow control variables for a “context” established for the block; a context memory for storing the variables; and a receive filter that updates flow control information for the block based on ACK packets received from the remote endpoint receiving the data packets.

Because the network adapter implements flow control for data blocks that it segments, intermediate ACK packets corresponding to that block can be intercepted by the adapter, before they pass to the host, conserving host resources. An added advantage is that the host CPU can offload data blocks larger than the remote endpoint'"'"'s receive window size, since the adapter can follow the transmit window and transmit packets at appropriate intervals. This further decreases load on the host CPU, decreases latency, and improves bandwidth utilization.

Claims

1. A method for operating a network adapter coupled to a host computer, the adapter having a host-resident driver and separate adapter hardware, the method comprising:

the network adapter driver accepting a request from a host-based transmission protocol layer to transmit a block of data to a remote endpoint;

the network adapter segmenting the block of data into multiple data packets;

the network adapter hardware transmitting the packets to the remote endpoint; and

the network adapter interpreting acknowledgment data, relevant to those packets, sent by the remote endpoint to the host-based transmission protocol layer, wherein interpreting comprises applying a filter to received acknowledgment packets, the filter applying a selection criteria to prevent at least some received acknowledgment packets addressed to the host-based transmission protocol layer from being forwarded to the host-based transmission protocol layer.

2. The method of claim 1, wherein applying the selection criteria comprises examining acknowledgment data piggybacked on data packets, before forwarding such data packets to the host-based transmission protocol layer.

3. The method of claim 1, further comprising resetting a retransmission time value when one of the received acknowledgment packets indicates that one or more of the data packets has been received by the remote endpoint.

4. The method of claim 1, wherein applying the selection criteria comprises examining the range of data values acknowledged by one of the acknowledgment packets and determining whether the host-based transmission protocol layer needs to be informed of the acknowledgment of that range of data values.

5. The method of claim 2, wherein interpreting comprises reading flow control variables from received acknowledgment packets relevant to the data packets.

6. The method of claim 1, wherein the block of data is larger than the window size of the remote endpoint at the time of the request, the method further comprising the network adapter hardware controlling the transmission timing of the multiple packets based on the remote endpoint'"'"'s window size and interpreted acknowledgment data.

7. The method of claim 6, further comprising, upon receiving the request from the host-based transport protocol layer to transmit a block of data to a remote endpoint, the network adapter hardware initializing a connection context corresponding to the request.

8. The method of claim 7, wherein controlling the transmission timing comprises:

periodically examining the remote endpoint'"'"'s window size and comparing it to the payload size of transmitted but unacknowledged data packets; and

building and transmitting another data packet for the connection context when the comparing step indicates that the remote endpoint can accept another data packet with at least a defined minimum payload size.

9. A method for operating a network adapter coupled to a host computer, the adapter having a host-resident driver and separate adapter hardware, the method comprising:

the network adapter driver accepting a request from a host-based transmission protocol layer to transmit a block of data to a remote endpoint, wherein the block of data is larger than the window size of the remote endpoint at the time of the request, the network adapter hardware initializing a connection context corresponding to the request;

the network adapter segmenting the block of data into multiple data packets;

the network adapter hardware transmitting the packets to the remote endpoint;

the network adapter hardware interpreting acknowledgment data, relevant to those packets, sent by the remote endpoint to the host-based transmission protocol layer; and

the network adapter hardware controlling the transmission timing of the multiple packets based on the remote endpoint'"'"'s window size and interpreted acknowledgment data;

wherein the network adapter is capable of accepting multiple requests and initializing a connection context for each request, the method further comprising interleaving service for the multiple contexts.

10. The method of claim 9, wherein interleaving service comprises serving the multiple contexts with a round-robin scheme, a priority-based scheme, or a combination of the two schemes.

11. The method of claim 9, further comprising:

establishing a retransmission time for the connection context; and

executing a retransmission procedure when an acknowledgment packet that acknowledges at least one of the transmitted data packets for that connection context is not received prior to the retransmission time.

12. The method of claim 11, wherein executing the retransmission procedure comprises aborting the connection context and reporting an error to the host-based transmission protocol layer.

13. The method of claim 11, wherein executing the retransmission procedure comprises retransmitting unacknowledged data packets for the block.

14. The method of claim 13, further comprising counting the number of times that retransmittal has occurred for the connection context, and aborting the context and reporting an error to the host-based transport protocol if the number of times exceeds a defined maximum.

15. A network adapter comprising:

a network interface;

a packet buffer memory to buffer packets for transmission over the network interface;

a context engine to establish and service a connection context requested by a host;

a packet engine to segment a data block associated with the connection context into multiple data packets and place these packets into the packet buffer memory; and

a receive filter to associate acknowledgment packets, received via the network interface, with the connection context;

wherein the context engine controls when the packet engine places data packets in the packet buffer memory, according to flow control information taken from the acknowledgment packets.

16. The network adapter of claim 15, wherein the context engine is integrated with the packet engine.

17. The network adapter of claim 15, wherein the receive filter is integrated with the context engine.

18. The network adapter of claim 15, further comprising a context memory to store variables pertinent to the context.

19. The network adapter of claim 18, wherein the context memory stores variables pertinent to multiple contexts, and wherein the context engine is capable of handling multiple open contexts.

20. The network adapter of claim 15, the receive filter comprising an acknowledgment packet trap having the capability to select at least one of the associated acknowledgment packets for the connection context and prevent that selected acknowledgment packet from being relayed from the network adapter to the host.

21. The network adapter of claim 15, further comprising a clock, the context engine referencing the clock to detect when an expected acknowledgment packet for the connection context is overdue.

22. The network adapter of claim 21, the context servicing engine comprising retransmission logic for instructing the packet engine to rebuild unacknowledged data packets for the connection context when the expected acknowledgment packet for that context is overdue.

23. The network adapter of claim 22, the context engine comprising retransmission counter logic to constrain the number of times that the retransmission logic can be used for the connection context.

24. A computer system comprising:

network adapter hardware having an offload operational mode wherein it segments a block of data into smaller data packets and configures the data packets for transmission under a network transport protocol, and then provides flow control for the data packets,

the network adapter hardware comprising a receive filter that, in the offload operational mode, interprets acknowledgment packets transmitted by a remote endpoint to the host network transport protocol in order to provide flow control for the packets, wherein the receive filter traps at least some of the acknowledgment packets, preventing their communication to the host network transport protocol; and

a host processor in communication with the network adapter hardware, the host processor being software configured such that, for a given block of data, the network transport protocol can either be executed on the host processor, or network transport protocol segmentation and flow control for the block can be handed off to the network adapter hardware for processing in the offload operational mode.

25. An article of manufacture comprising a computer-readable medium containing a driver program for a network adapter, the driver program causing a processor to execute:

a packet segmentation offload scheduler that (a) accepts requests from a higher-level protocol to segment a data block and temporarily handle flow control for that block, and (b) schedules those requests, including flow control, to network adapter hardware controlled by the driver program, wherein the packet segmentation offload scheduler tracks the number of requests currently being handled by the network adapter, and rejects further requests from the higher-level protocol when the number of current requests exceeds a preset maximum; and

a packet segmentation offload status reporter for communicating the status of accepted requests to the higher-level protocol.