Sharing a network interface card among multiple hosts

• US 20030200315A1
• Filed: 04/23/2002
• Published: 10/23/2003
• Est. Priority Date: 04/23/2002
• Status: Active Grant

• Current Assignee: Mellanox Technologies Limited (NVIDIA Corporation)
• Original Assignee: Mellanox Technologies Limited (NVIDIA Corporation)
• Inventors: Goldenberg, Dror, Crupnicoff, Diego, Kagan, Michael, Bloch, Gil, Stoler, Gil
• Granted Patent: US 7,245,627 B2
• Time in Patent Office: Days
• US Class Current: 709/225
• CPC Class Codes: H04L 47/10 Flow control; Congestion control
• CPC Class Codes: H04L 47/125 by balancing the load, e.g. traffic engineering
• CPC Class Codes: H04L 47/43 Assembling or disassembling of packets, e.g. segmentation and reassembly [SAR]
• CPC Class Codes: H04L 49/25 Routing or path finding in a switch fabric
• CPC Class Codes: H04L 49/30 Peripheral units, e.g. input or output ports
• CPC Class Codes: H04L 49/3009 Header conversion, routing tables or routing tags
• CPC Class Codes: H04L 49/351 for local area network [LAN], e.g. Ethernet switches

Abstract

A network interface device includes a fabric interface, adapted to exchange messages over a switch fabric with a plurality of host processors, the messages containing data, and a network interface, including one or more ports adapted to be coupled to a network external to the switch fabric. Message processing circuitry is coupled between the fabric interface and the network interface, so as to enable at least first and second host processors among the plurality of the host processors to use a single one of the ports substantially simultaneously so as to transmit and receive frames containing the data over the network.

Claims

1. A network interface device, comprising:

a fabric interface, adapted to exchange messages over a switch fabric with a plurality of host processors, the messages containing data;

a network interface, comprising one or more ports adapted to be coupled to a network external to the switch fabric; and

message processing circuitry, coupled between the fabric interface and the network interface, so as to enable at least first and second host processors among the plurality of the host processors to use a single one of the ports substantially simultaneously so as to transmit and receive frames containing the data over the network.

2. A device according to claim 1, wherein the fabric interface comprises a target channel adapter, and wherein the messages comprise data packets.

3. A device according to claim 1, wherein the network comprises an Ethernet network, and the frames comprise Ethernet frames.

4. A device according to claim 1, wherein each of the host processors is assigned to a respective one of the ports, and transmits and receives the frames over the one of the ports to which it is assigned.

5. A device according to claim 1, wherein at least one of the host processors is assigned to transmit and receive the frames over two or more of the ports simultaneously.

6. A device according to claim 1, wherein the messages comprise outgoing messages sent by the host processors over the switch fabric, and wherein the message processing circuitry is adapted to extract the data from the outgoing messages and to generate the frames containing the extracted data for transmission over the network.

7. A device according to claim 6, wherein the message processing circuitry is adapted to assign respective service levels to the host processors, and to determine an order of processing the outgoing messages responsive to the service levels.

8. A device according to claim 7, wherein the fabric interface is adapted to exchange the messages with the host processors over respective transport connections between the host processors and the fabric interface, and wherein the message processing circuitry is adapted to assign each of the transport connections to a respective one of the service levels.

9. A device according to claim 7, wherein the message processing circuitry comprises a plurality of scheduling queues having respective processing priorities, and is adapted to place each of the outgoing messages in a respective one of the scheduling queues responsive to the service levels assigned to the host processors that sent the messages.

10. A device according to claim 6, wherein each of the host processors is assigned to a respective one of the ports of the network interface, and wherein the message processing circuitry comprises one or more scheduling queues for each of the one or more ports, and is adapted to place the outgoing messages in the scheduling queues depending on the respective ports to which the host processors that sent the messages are assigned.

11. A device according to claim 6, wherein the message processing circuitry is adapted to determine whether any of the frames generated for transmission over the network in response to one of the outgoing messages from the first host processor are destined to be received by the second host processor and, if so, to loop the data in the frames destined to be received by the second host processor back to the fabric interface, to be conveyed over the switch fabric to the second host processor without being transmitted over the network.

12. A device according to claim 11, wherein the message processing circuitry comprises a destination address filter for each of the ports, which filter is adapted to process destination addresses of the frames generated for transmission over the network so as to determine whether any of the frames are destined to be received by any of the host processors, including the second host processor.

13. A device according to claim 11, wherein the frames for transmission over the network comprise a multicast frame, which is destined to be received by both a remote recipient on the network and by the second host processor, and wherein the message processing circuitry is adapted both to generate the multicast frame for transmission over the network to the remote recipient and to loop the data in the multicast frame back to be conveyed to the second host processor.

14. A device according to claim 13, wherein the message processing circuitry is controllable so that the data in the multicast frame are looped back to the second host processor, while the transmission of the multicast frame over the network is blocked.

15. A device according to claim 1, wherein the messages comprise incoming messages sent from the fabric interface to the host processors, and wherein the message processing circuitry is adapted to extract the data from the frames received over the network and to generate the incoming messages containing the extracted data to be sent to the host processors.

16. A device according to claim 15, wherein the message processing circuitry comprises a destination address filter, which is adapted to process destination addresses of the frames received over the network so as to indicate which of the host processors should receive the incoming messages over the switch fabric.

17. A device according to claim 16, wherein the fabric interface is adapted to exchange the messages with the host processors over respective transport connections between the host processors and the fabric interface, and wherein the message processing circuitry is adapted to determine the transport connections over which to send the incoming messages to the host processors indicated by the destination address filter.

18. A device according to claim 16, wherein the destination addresses processed by the destination address filter comprise at least one of a Media Access Control (MAC) address and a Virtual Local Area Network (VLAN) tag.

19. A device according to claim 16, wherein the frames received over the network comprise a multicast frame, and wherein the destination address filter is adapted to indicate that one or more of the hosts should receive the incoming messages containing the data extracted from the multicast frame.

20. A device according to claim 16, wherein the messages further comprise outgoing messages sent by the host processors over the switch fabric, and wherein the message processing circuitry is adapted to extract the data from the outgoing messages and to generate the frames containing the data extracted from the outgoing messages for transmission over the network, and wherein the destination address filter is adapted to process the destination addresses of the generated frames so as to determine whether any of the frames are destined to be received by any of the host processors and, if so, to loop the data in the frames back to the fabric interface, to be conveyed over the switch fabric to the host processors that are destined to receive them.

21. A device according to claim 15, wherein the message processing circuitry is adapted to extract the data from two or more of the frames received over the network and destined for a given one of the host processors, and to generate one of the incoming messages containing the data extracted from the two or more of the frames.

22. A device according to claim 1, wherein the message processing circuitry comprises a plurality of counters, which are adapted to collect performance statistics individually for each of the first and second host processors regarding the frames transmitted and received using the single one of the ports.

23. A method for interfacing a plurality of host processors to a network, comprising:

providing a network interface device having one or more ports connected to a network;

coupling the host processors to exchange messages containing data over a switch fabric with the network interface device, such that at least first and second host processors among the plurality of the host processors are assigned to use a single one of the ports of the network interface device; and

processing the messages exchanged with the host processors so as to enable the first and second host processors to use the single one of the ports substantially simultaneously to transmit and receive frames containing the data over the network.

24. A method according to claim 23, wherein coupling the host processors comprises coupling the host processors and the network interface device to the switch fabric using respective channel adapters, and wherein the messages comprise data packets.

25. A method according to claim 23, wherein the network comprises an Ethernet network, and the frames comprise Ethernet frames.

26. A method according to claim 23, wherein coupling the host processors comprises assigning each of the host processors to a respective one of the ports, so as to transmit and receive the frames over the one of the ports to which it is assigned.

27. A method according to claim 23, wherein coupling the host processors comprises assigning at least one of the host processors to transmit and receive the frames over two or more of the ports simultaneously.

28. A method according to claim 23, wherein processing the messages comprises processing outgoing messages sent by the host processors over the switch fabric, so as to extract the data from the outgoing messages and to generate the frames containing the extracted data for transmission over the network.

29. A method according to claim 28, wherein processing the outgoing messages comprises assigning respective service levels to the host processors, and determining an order of processing the outgoing messages responsive to the service levels.

30. A method according to claim 29, wherein coupling the host processors to exchange the messages comprises exchanging the messages over respective transport connections between the host processors and the network interface device, and wherein assigning the respective service levels comprises assigning each of the transport connections to a respective one of the service levels.

31. A method according to claim 29, wherein assigning the respective services levels comprises placing each of the outgoing messages in a respective one of a plurality of scheduling queues having respective processing priorities, responsive to the service levels assigned to the host processors that sent the messages.

32. A method according to claim 28, wherein coupling the host processors comprises assigning each of the host processors to a respective one of the ports so as to transmit the frames containing the extracted data through the one of the ports to which it is assigned, and wherein processing the messages comprises providing one or more scheduling queues for each of the one or more ports, and placing each of the outgoing messages in a respective one of the scheduling queues depending on the respective one of the ports to which the host processor that sent the message is assigned.

33. A method according to claim 32, wherein coupling the host processors comprises opening one or more respective transport service instance over which to exchange the messages with each of the host processors, and wherein placing each of the outgoing messages in the respective one of the scheduling queues further comprises selecting the respective one of the scheduling queues for each of the outgoing messages responsive to the transport service instances over which the messages are sent by the host processors.

34. A method according to claim 28, wherein processing the messages comprises determining whether any of the frames generated for transmission over the network in response to one of the outgoing messages from the first host processor are destined to be received by the second host processor and, if so, looping the data in the frames destined to be received by the second host processor back to the fabric interface, to be conveyed over the switch fabric to the second host processor without being transmitted over the network.

35. A method according to claim 34, wherein determining whether any of the frames are destined to be received by the second host processor comprises providing a destination address filter for each of the ports, and processing destination addresses of the frames generated for transmission over the network using the filter so as to determine whether any of the frames are destined to be received by any of the host processors, including the second host processor.

36. A method according to claim 34, wherein processing the outgoing messages comprises determining that a multicast frame is destined to be received by both a remote recipient on the network and by the second host processor, and generating the multicast frame for transmission over the network to the remote recipient while looping the data in the multicast frame back to be conveyed to the second host processor.

37. A method according to claim 36, wherein looping the data back comprises determining, responsive to a predefined policy, that the data in the multicast frame are to be looped back to the second host processor, while blocking the transmission of the multicast frame over the network.

38. A method according to claim 23, wherein processing the messages comprises processing the frames received over the network so as to extract the data therefrom, and wherein the messages comprise incoming messages containing the extracted data, sent from the network interface device to the host processors.

39. A method according to claim 38, wherein processing the messages comprises providing a destination address filter, and processing destination addresses of the frames received over the network using the filter so as to indicate which of the host processors should receive the incoming messages over the switch fabric.

40. A method according to claim 39, wherein coupling the host processors to exchange the messages comprises exchanging the messages with the host processors over respective transport connections between the host processors and the network interface device, and processing the destination addresses comprises determining the transport connections over which to send the incoming messages to the host processors indicated by the destination address filter.

41. A method according to claim 39, wherein processing the destination addresses comprises processing at least one of a Media Access Control (MAC) address and a Virtual Local Area Network (VLAN) tag.

42. A method according to claim 39, wherein processing the frames received over the network comprises receiving a multicast frame, and wherein processing the destination addresses comprises determining, using the destination address filter, that one or more of the hosts should receive the incoming messages containing the data extracted from the multicast frame.

43. A method according to claim 39, wherein processing the messages further comprises processing outgoing messages sent by the host processors over the switch fabric, so as to extract the data from the outgoing messages and to generate the frames containing the data extracted from the outgoing messages for transmission over the network, and wherein processing the outgoing messages comprises processing the destination addresses of the generated frames using the destination address filter so as to determine whether any of the frames are destined to be received by any of the host processors and, if so, looping the data in the frames back to be conveyed over the switch fabric to the host processors that are destined to receive them.

44. A method according to claim 38, wherein processing the frames comprises extracting the data from two or more of the frames received over the network that are destined for a given one of the host processors, and generating one of the incoming messages so as to contain the data extracted from the two or more of the frames.

45. A method according to claim 23, wherein the processing the messages comprises collecting performance statistics individually for each of the first and second host processors regarding the frames transmitted and received using the single one of the ports.

46. A network interface device, comprising:

a fabric interface, adapted to exchange messages over a switch fabric with a plurality of host processors, including first and second host processors, the messages containing data and including outgoing messages sent by the host processors over the switch fabric;

a network interface, adapted to be coupled to a network external to the switch fabric; and

message processing circuitry, coupled between the fabric interface and the network interface, and adapted to extract the data from the outgoing messages and to generate frames containing the extracted data for transmission over the network via the network interface, and further adapted to determine whether any of the frames generated in response to one of the outgoing messages from the first host processor are destined to be received by the second host processor and, if so, to loop the data in the frames destined to be received by the second host processor back to the fabric interface, to be conveyed over the switch fabric to the second host processor without being transmitted over the network.

47. A device according to claim 46, wherein the message processing circuitry comprises a destination address filter, which is adapted to process destination addresses of the frames generated for transmission over the network so as to determine whether any of the frames are destined to be received by any of the host processors, including the second host processor.

48. A device according to claim 46, wherein the frames for transmission over the network comprise a multicast frame, which is destined to be received by both a remote recipient on the network and by the second host processor, and wherein the message processing circuitry is adapted both to generate the multicast frame for transmission over the network to the remote recipient and to loop the data in the multicast frame back to be conveyed to the second host processor.

49. A device according to claim 48, wherein the message processing circuitry is controllable so that the data in the multicast frame are looped back to the second host processor, while the transmission of the multicast frame over the network is blocked.

50. A network interface device, comprising:

a fabric interface, adapted to exchange messages over a switch fabric with a plurality of host processors, the messages containing data;

a network interface, adapted to be coupled to a network external to the switch fabric; and

message processing circuitry, coupled between the fabric interface and the network interface, so as to enable at least first and second host processors among the plurality of the host processors to transmit and receive frames containing the data over the network via the network interface, the message processing circuitry comprising a plurality of counters, which are adapted to collect performance statistics individually for each of the first and second host processors regarding the frames transmitted and received via the network interface.

51. A method for interfacing a plurality of host processors to a network, comprising:

connecting a network interface device to a network;

coupling the host processors, including at least first and second host processors among the plurality of the host processors, to exchange messages containing data over a switch fabric with the network interface device, the messages including outgoing messages sent by the host processors over the switch fabric;

processing the messages so as to extract the data from the outgoing messages and to generate frames containing the extracted data for transmission over the network via the network interface device; and

determining whether any of the frames generated for transmission over the network in response to one of the outgoing messages from the first host processor are destined to be received by the second host processor and, if so, looping the data in the frames destined to be received by the second host processor back to the fabric interface, to be conveyed over the switch fabric to the second host processor without being transmitted over the network.

52. A method according to claim 51, wherein determining whether any of the frames are destined to be received by the second host processor comprises providing a destination address filter for each of the ports, and processing destination addresses of the frames generated for transmission over the network using the filter so as to determine whether any of the frames are destined to be received by any of the host processors, including the second host processor.

53. A method according to claim 51, wherein processing the messages comprises determining that a multicast frame is destined to be received by both a remote recipient on the network and by the second host processor, and generating the multicast frame for transmission over the network to the remote recipient while looping the data in the multicast frame back to be conveyed to the second host processor.

54. A method according to claim 53, wherein processing the messages further comprises determining, responsive to a predefined policy, that the data in the multicast frame are to be looped back to the second host processor, while blocking the transmission of the multicast frame over the network.

55. A method for interfacing a plurality of host processors to a network, comprising:

connecting a network interface device to a network;

coupling the host processors to exchange messages containing data over a switch fabric with the network interface device;

processing the messages exchanged with the host processors so as to enable the first and second host processors to transmit and receive frames containing the data over the network via the network interface device; and

collecting performance statistics at the network interface device individually for each of the first and second host processors regarding the frames transmitted and received via the network interface device.