Data affinity based scheme for mapping connections to CPUs in I/O adapter

• US 8,949,472 B2
• Filed: 09/10/2008
• Issued: 02/03/2015
• Est. Priority Date: 09/10/2008
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Shalev, Leah, Pinter, Shlomit S.
• Primary Examiner(s): Donabed, Ninos
• Assistant Examiner(s): RICHARDSON, THOMAS W
• Application Number:
  US12/207,561
• Publication Number:
  US 20100064286A1
• Time in Patent Office: 2,337 Days
• Field of Search: 709/250, 718/102
• US Class Current: 709/250
• CPC Class Codes: G06F 15/17318 Parallel communications techniques, e.g. gather, scatter, reduce, roadcast, multicast, all to all
• CPC Class Codes: H04L 12/185 with management of multicast group membership
• CPC Class Codes: H04L 12/1881 with schedule organisation, e.g. priority, sequence management
• CPC Class Codes: H04L 69/32 Architecture of open systems interconnection [OSI] 7-layer type protocol stacks, e.g. the interfaces between the data link level and the physical level
• CPC Class Codes: H04L 69/326 in the transport layer [OSI layer 4] H04L69/16 takes precedence

Abstract

A method, system and computer program product is disclosed for scheduling data packets in a multi-processor system comprising a plurality of processor units and a multitude of multicast groups. The method comprises associating one of the processor units with each of the multicast groups, receiving a multitude of data packets from the multicast groups, and scheduling all of the data packets received from each of the multicast groups for processing by the one of the processor units associated with said each of the multicast groups. In one embodiment, scheduling is based on affinity of both transmit and received processing for multiple connections to a processor unit. In another embodiment, a system call is provided for transmitting the same data over multiple sockets. Additional system calls may be used for building multicast group socket lists.

Claims

1. A method of scheduling packets in a multi-processor system comprising a plurality of processor units, a plurality of caches, wherein each of a distinct one of the processor units is associated with each of a distinct one of the caches, a plurality of multicast groups, wherein each multicast group includes a plurality of members, and a single network adapter, the method comprising:

associating a distinct one of the processor units with each of a distinct one of the multicast groups;

receiving a multitude of data packets from the multicast groups via the single network adapter;

scheduling all of the data packets received from each of the multicast groups for processing by the one of the processor units associated with said each of the multicast groups;

handling all of a plurality of connections belonging to a given one of the multicast groups by the same one of the processor units; and

holding a single copy of each packet in a respective one of the caches associated with the processor unit that is processing the respective packet to be sent via the single network adapter to all members of a respective multicast group;

wherein the same data is transmitted by a single local thread to all members of a multicast group;

wherein a plurality of connections are provided for each of the processor units, and the scheduling all of the data packets received from each of the multicast groups for processing by the one of the processor units associated with said each of the multicast groups includes mapping all of the data packets received from each of the multicast groups to the connections provided for the processor unit associated with said each of the multicast groups.

2. A method according to claim 1, wherein all of the plurality of connections belonging to a given one of the multicast groups, including both transmit and receive, are handled by the same one of the processor units.

3. A method according to claim 1, wherein the scheduling all of the data packets received from each of the multicast groups for processing by the one of the processor units associated with said each of the multicast groups includes recognizing each multicast group, and mapping all of the members of each multicast group to connections on the same processor unit.

4. A method according to claim 1, wherein the scheduling all of the data packets received from each of the multicast groups for processing by the one of the processor units associated with said each of the multicast groups includes identifying the multicast group from which each of the data packets is received.

5. A method according to claim 1, wherein the multi-processor system includes an I/O adapter for receiving the data packets, and the scheduling is done by said I/O adapter.

6. A method according to claim 1, wherein the scheduling is based on a system call for transmitting the same data over multiple sockets.

7. A method according to claim 1, wherein the scheduling of packets is performed in the context of an application using a publish/subscribe protocol.

8. A scheduling system for scheduling packets in a multi-processor system comprising a plurality of processor units, a plurality of caches, wherein each of a distinct one of the processor units is associated with each of a distinct one of the caches, a plurality of multicast groups, wherein each multicast group includes a plurality of members, and a single network adapter, the scheduling system comprising one or more of said processor units configured for:

associating a distinct one of the processor units with each of a distinct one of the multicast groups;

receiving a multitude of data packets from the multicast groups via the single network adapter;

scheduling all of the data packets received from each of the multicast groups for processing by the one of the processor units associated with said each of the multicast groups;

handling all of a plurality of connections belonging to a given one of the multicast groups by the same one of the processor units; and

holding a single copy of each packet in a respective one of the caches associated with the processor unit that is processing the respective packet to be sent via the single network adapter to all members of a respective multicast group;

wherein the same data is transmitted by a single local thread to all members of a multicast group;

wherein a plurality of connections are provided for each of the processor units, and the scheduling all of the data packets received from each of the multicast groups for processing by the one of the processor units associated with said each of the multicast groups includes mapping all of the data packets received from each of the multicast groups to the connections provided for the processor unit associated with said each of the multicast groups.

9. A scheduling system according to claim 8, wherein all of the plurality of connections belonging to a given one of the multicast groups, including both transmit and receive, are handled by the same one of the processor units.

10. A scheduling system according to claim 8, wherein the scheduling includes the step of recognizing each multicast group, and mapping all of the members of each multicast group to connections on the same processor unit.

11. A scheduling system according to claim 8, wherein the scheduling all of the data packets includes identifying the multicast group from which each of the data packets is received.

12. A scheduling system according to claim 8, wherein the multi-processor system includes an I/O adapter for receiving the data packets, and the scheduling is done by said I/O adapter.

13. A scheduling system according to claim 8, wherein the scheduling of packets is performed in the context of an application using a publish/subscribe protocol.

14. A computer program product for scheduling packets in a multi-processor system comprising a plurality of processor units, a plurality of caches, wherein each of a distinct one of the processor units is associated with each of a distinct one of the caches, a plurality of multicast groups, wherein each multicast group includes a plurality of members, and a single network adapter, the computer program product comprising:

a computer usable hardware device having computer usable program code embodied therewith, the computer usable program code comprising;

computer usable program code configured for;

associating a distinct one of the processor units with each of a distinct one of the multicast groups;

receiving a multitude of data packets from the multicast groups via the single network adapter;

scheduling all of the data packets received from each of the multicast groups for processing by the one of the processor units associated with said each of the multicast groups;

handling all of a plurality of connections belonging to a given one of the multicast groups by the same one of the processor units; and

holding a single copy of each packet in a respective one of the caches associated with the processor unit that is processing the respective packet to be sent via the single network adapter to all members of a respective multicast group;

wherein the same data is transmitted by a single local thread to all members of a multicast group,

wherein a plurality of connections are provided for each of the processor units, and the scheduling all of the data packets received from each of the multicast groups for processing by the one of the processor units associated with said each of the multicast groups includes mapping all of the data packets received from each of the multicast groups to the connections provided for the processor unit associated with said each of the multicast groups.

15. A computer program product according to claim 14, wherein all of the plurality of connections belonging to a given one of the multicast groups, including both transmit and receive, are handled by the same one of the processor units.

16. A computer program product according to claim 14, wherein the scheduling all of the data packets received from each of the multicast groups for processing by the one of the processor units associated with said each of the multicast groups is based on a system call for transmitting the same data over multiple sockets.

17. A computer program product according to claim 14, wherein the scheduling of packets is performed in the context of an application using a publish/subscribe protocol.