Method and apparatus for dynamic detection and routing of non-uniform traffic in parallel buffered multistage interconnection networks

• US 5,274,782 A
• Filed: 08/27/1990
• Issued: 12/28/1993
• Est. Priority Date: 08/27/1990
• Status: Expired due to Term

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Chalasani, Suresh, Varma, Anujan M.
• Primary Examiner(s): RUDOLPH, REBECCA
• Application Number:
  US07/573,610
• Time in Patent Office: 1,219 Days
• Field of Search: 364/200 MS File, 364/900 MS File, 395/425, 395/200, 395/800, 395/325, 370/60, 370/94.1
• US Class Current: 710/317
• CPC Class Codes: G06F 15/17393 having multistage networks, e.g. broadcasting scattering, gathering, hot spot contention, combining/decombining

Abstract

A method and apparatus for routing processor-memory data traffic in a shared-memory multiprocessor computer system employs an interconnection network including two buffered multistage switching networks. Each of these networks can be used to route the data from any processing element to any memory element. Depending on the nature of the processor-memory traffic, two distinct routing schemes are used to distribute the traffic among the two networks. The first method distributes the memory accesses evenly among the two networks and maximizes performance when the memory accesses are uniformly distributed among the memory modules. However, when the traffic is highly non-uniform, a second routing method is used to confine the non-uniform part of the traffic to one network and the remaining part to the other network. The routing method is selected based on the prevailing traffic conditions. A distributed feedback mechanism detects the change in traffic conditions and changes the routing method accordingly. A traffic monitoring circuit within each memory module monitors the traffic into the memory module continuously and senses a change in the traffic condition. The condition is conveyed to the processing elements by means of a status flag associated with each response message from the memory module to processing elements. The processing elements respond to a change in traffic condition by switching to the alternate routing method.

Claims

1. Apparatus for mounting processor-memory traffic in a shared-memory multiprocessor computer system of the type comprising a plurality of processing elements and a plurality of memory modules, comprising:

an interconnection system having at least two multistage switching networks, said at least two multistage switching networks each interconnecting said plurality of processing elements and said plurality of memory modules;

means for detecting traffic non-uniformities in each memory module, said detecting means counting a number of requests to a memory module during a predetermined number of cycles and comparing a count to first and second predetermined thresholds; and

feedback means to notifying said plurality of processing elements as to a status of each memory module; and

means within each processing element responsive to said feedback means for selecting one of said at least two networks for routing a request based on two different routing methods, a first of said routing methods suited to uniform traffic conditions and a second of said routing methods suited to non-uniform traffic conditions, said means for selecting comprising;

a memory-status table addressed by a memory address from one of said processing elements, said memory-status table providing a first output indicating a status of an addressed memory module;

a selector circuit responsive to said first output of said memory-status table and generating a select output;

packet router means responsive to said select output for routing a memory request from said one of said processing elements to one of said switching networks; and

counter means responsive to said feedback means for maintaining a count of memory modules having a non-uniform traffic status.

2. The apparatus for routing processor-memory traffic in a shared-memory multiprocessor computer system recited in claim 1 wherein said means for selecting further includes means for preventing memory requests from a corresponding one of said processing elements from getting out of order due to said requests being routed through different ones of said at least two switching networks.

3. The apparatus for routing processor-memory traffic in a shared-memory multiprocessor system recited in claim 2 wherein said means for preventing memory requests from getting out of order comprises:

an outstanding-request table addressed by said memory address from said one of said processing elements, said outstanding-request table providing a first output indicating whether an addressed memory module has been previously addressed but no response has been received from said addressed memory module and a second output identifying the switching network which a previous request was routed; and

comparing means in said selector circuit enables by said first output of said outstanding-request table for comparing said second output with said select output to determine if there is a match and, if so, generating a hold packet output to said processing element.

4. The apparatus for routing processor-memory traffic in a shared-memory multiprocessor computer system recited in claim 1 wherein said means for detection of traffic non-uniformities includes detecting means in each of said memory modules for designating the memory modules as hot or cool based on monitoring traffic to the module.

5. The apparatus for routing processor-memory traffic in a shared-memory multiprocessor computer system as recited in claim 4 wherein said detecting means comprises:

windowing means for counting the number of requests to a memory module during a predetermined number of cycles;

comparing means connected to said windowing means for comparing a count in said windowing means to first and second predetermined thresholds; and

flag means set by said count exceeding said first threshold and reset by said count going below said second threshold.

6. The apparatus for routing processor-memory traffic in a shared-memory multiprocessor computer system as recited in claim 5 wherein said first and second predetermined thresholds are set by an operating system of said multiprocessor computer system.

7. The apparatus for routing processor-memory traffic in a shared-memory multiprocessor computer system as recited in claim 5 further comprising programmable means in said windowing means for charging the number of cycles during which memory requests are counted.

8. The apparatus for routing processor-memory traffic in a shared-memory multiprocessor computer system as recited in claim 5 wherein said interconnection system includes a response network for routing response data from said memory modules to said processing elements and said feedback means includes packet formatter means at each of said modules connected to said response network and receiving said response data and responsive to said flag means for setting a flag indicating traffic non-uniformities at the module.

9. The apparatus for routing processor-memory traffic in a shared-memory multiprocessor system recited in claim 8 wherein said means for selecting comprises:

a memory-status table addressed by a memory address from said one of said processing elements, said memory-status table providing a first output indicating a status of an addressed memory module;

a selector circuit responsive to said first output of said memory-status table and generating a select output; and

packet router means responsive to said select output for routing a memory request from said one of said processing elements to one of said switching networks.

10. The apparatus for routing processor-memory traffic in a shared-memory multiprocessor computer system recited in claim 9 wherein said means for selecting further includes means for preventing memory requests from a corresponding one of said processing elements from getting out of order due to said requests being routed through different ones of said at least two switching networks.

11. The apparatus for routing processor-memory traffic in a shared-memory multiprocessor computer system recited in claim 10 wherein said means for preventing memory requests from getting out of order comprises:

an outstanding-request table addressed by said memory address from said one of said processing elements, said outstanding-request table providing a first output indicating whether an addressed memory module has been previously addressed but no response has been received from said addressed memory module and a second output identifying the switching network over which the previous request was routed; and

comparing means in said selector circuit enabled by said first output of said outstanding-request table for comparing said second output with select output to determine if there is a match and, if so, generating a hold packet output to said processing element.

12. The apparatus for routing processor-memory traffic in a shared memory multiprocessor computer system recited in claim 11 wherein said means within each processing element further comprises a packet deformatter providing as a first output an address of a responding memory module and as a second output the status of the responding memory module and wherein said means for selecting further comprises counter means responsive to said second output of said deformatter for maintaining a count of memory modules having a non-uniform traffic status, said first and second outputs being supplied to said memory-status table for updating the memory-status table and said first output being supplied to said outstanding-request table for updating the outstanding-request table.