System and method for implementing multi-pathing data transfers in a system area network

• US 6,493,343 B1
• Filed: 12/30/1998
• Issued: 12/10/2002
• Est. Priority Date: 01/07/1998
• Status: Expired due to Term

• Current Assignee: Hewlett Packard Enterprise Development LP (Hewlett-Packard Enterprise Company)
• Original Assignee: Compaq Computer Corporation (HP Inc.)
• Inventors: Garcia, David J., Larson, Richard O.
• Primary Examiner(s): Yao, Kwang Bin
• Assistant Examiner(s): Harper, Kevin C.
• Application Number:
  US09/224,114
• Time in Patent Office: 1,441 Days
• Field of Search: 370/464, 370/474, 370/389, 370/394, 370/400, 370/419, 370/235-236, 709/213, 709/217-219, 709/201, 709/203
• US Class Current: 370/394
• CPC Class Codes: G06F 12/023 Free address space management
• CPC Class Codes: G06F 12/0292 using tables or multilevel address translation means G06F12/023 takes precedence; address translation in virtual memory systems G06F12/10
• CPC Class Codes: G06F 12/1081 for peripheral access to main memory, e.g. direct memory access [DMA]
• CPC Class Codes: H04L 1/16 in which the return channel carries supervisory signals, e.g. repetition request signals
• CPC Class Codes: H04L 1/188 Time-out mechanisms
• CPC Class Codes: H04L 1/22 using redundant apparatus to increase reliability
• CPC Class Codes: H04L 49/30 Peripheral units, e.g. input or output ports
• CPC Class Codes: H04L 49/557 Error correction, e.g. fault recovery or fault tolerance
• CPC Class Codes: H04L 49/90 Buffering arrangements
• CPC Class Codes: H04L 69/26 Special purpose or proprietary protocols or architectures network applications for proprietary or special purpose networking environments H04L67/12
• CPC Class Codes: H04L 69/324 in the data link layer [OSI layer 2], e.g. HDLC
• CPC Class Codes: H04L 69/326 in the transport layer [OSI layer 4] H04L69/16 takes precedence
• CPC Class Codes: H04L 9/40 Network security protocols

Abstract

A system and method for facilitating both in-order and out-of-order packet reception in a SAN includes requester and responder nodes that maintain local copies of a message sequence number. Each request packet includes an ordering field specifying whether the packets must be received in-order. The request node includes a copy of the local sequence number in each packet transmitted and increments its local copy of the sequence number only for packets that must be received in order. The responder node includes the received message sequence number in all response packets and increments its local copy of the message sequence number only if the ordering field specifies that the packets must be received in order.

Claims

1. In a system area network (SAN) including multiple nodes coupled by a network fabric, a system for transferring data between a requestor node and a responder node, with the SAN implementing data transfers as a sequence of request/response packet pairs, a sub-system for managing ordered transactions requiring strict ordering of packet reception and remote direct memory access (RDMA) that allow out-of-order receipt of packets, said sub-system comprising:

a first network interface card, coupled to said requestor node, including request logic for maintaining a request sequence number for each transaction and transaction logic for translating ordered transaction requests or RDMA transaction requests into sequences of request packets and for including a packet sequence number equal to the request sequence number in each packet, with each packet including a packet order field having either a first value specifying that packets must be received in order or having a second value specifying that packets can be received out-of-order, and where, for an ordered transaction, the transaction logic sets said packet ordering field in each packet sent to the first value and increments the request sequence number for each packet sent, and where, for an unordered transaction, transaction logic sets said packet ordering field in each packet sent to the second value and does not increment the request sequence number for each packet sent; and

a second network interface card, coupled to said responder node and coupled to receive request packets sent from said first network interface card, said second NIC including response logic for generating and sending a response packet for each received request packet if the packet sequence number is equal to the response sequence number, with said response logic maintaining a response sequence number, and with said response logic copying the packet sequence number from the received request packet into a corresponding response packet, incrementing the local value of the sequence number subsequent to sending a response packet if the packet sequence number is equal to the response sequence number and the packet ordering field of the corresponding received request packet is equal to the first value, and for not incrementing the local value of the sequence number if the packet ordering field is equal to the second value.

2. The sub-system of claim 1 wherein each network interface card has two ports, the first network interface card and the second network interface card being coupled by a plurality of paths between the ports with said packets having a packet order field equal to the first value transmitted only over a single path and packets having a packet order field equal to the second value transmitted over a plurality of paths.

3. The sub-system of claim 1 wherein a plurality of packets may be transmitted from the requestor prior to receiving a response packet for any of the packets transmitted.

4. The sub-system of claim 1 wherein the local copies of the sequence numbers are initialized to a common value prior to data transfer between the requestor and responder nodes.

5. In a system area network (SAN) including multiple nodes coupled by a network fabric, a system for transferring data between a requestor node and a responder node, with the SAN implementing data transfers as a sequence of request/response packet pairs between a requestor and a responder, a method for managing ordered transactions requiring strict ordering of packet reception and remote direct memory access (RDMA) that allow out-of-order receipt of packets, said method comprising the steps of:

at said requestor;

maintaining a request sequence number for each transaction;

translating ordered transaction requests or RDMA transaction requests into sequences of request packets;

including a packet sequence number equal to the request sequence number in each packet, with each packet including a packet order field having either a first value specifying that packets must be received in order or having a second value specifying that packets can be received out-of-order, and, for an ordered transaction;

setting said packet ordering field in each packet sent to the first value; and

incrementing the request sequence number for each packet sent,

and, for an unordered transaction;

setting said packet ordering field in each packet sent to the second value and

not incrementing the request sequence number for each packet sent;

at said responder;

generating and sending a response packet for each received request packet;

maintaining a response sequence number;

copying the packet sequence number from the received request packet into a corresponding response packet,

incrementing the local value of the sequence number subsequent to sending a response packet if the packet sequence number is equal to the response sequence number and the packet ordering field of the corresponding received request packet is equal to the first value, and

not incrementing the local value of the sequence number if the packet ordering field is equal to the second value.