Multi processor sorting network for sorting while transmitting concurrently presented messages by message content to deliver a highest priority message

• US 5,276,899 A
• Filed: 08/10/1990
• Issued: 01/04/1994
• Est. Priority Date: 04/01/1981
• Status: Expired due to Term

• Current Assignee: Teradata US, Inc. (Teradata Corporation)
• Original Assignee: Teredata US, Inc.
• Inventors: Neches, Philip M.
• Primary Examiner(s): Lee, Thomas C.
• Assistant Examiner(s): Harrity, Paul
• Application Number:
  US07/565,818
• Time in Patent Office: 1,243 Days
• Field of Search: 395/800, 395/200, 395/325, 395/650, 340/825.02, 340/825.5, 370/60.60.1, 370/61
• US Class Current: 709/240
• CPC Class Codes: G06F 11/202 where processing functionality is redundant redundant communication control functionality G06F11/2005, redundant storage control functionality G06F11/2089
• CPC Class Codes: G06F 11/2094 Redundant storage or storage space G06F11/2056 takes precedence
• CPC Class Codes: G06F 15/161 Computing infrastructure, e.g. computer clusters, blade chassis or hardware partitioning casings, cabinets, racks or drawers for data centers H05K5/00
• CPC Class Codes: G06F 15/17368 non hierarchical topologies
• CPC Class Codes: G06F 16/10 File systems; File servers
• CPC Class Codes: Y10S 707/99939 Privileged access
• CPC Class Codes: Y10S 707/99945 Object-oriented database structure processing

Abstract

A multiprocessor system intercouples the processors with an active logic network having a plurality of priority determining nodes. Messages applied concurrently to the network in groups are sorted, using the data content of the messages, to a single or common priority message which is distributed to all the processors with a predetermined total network delay time. Losing messages are again retried concurrently in groups at a later time. Message routing is determined by local acceptance or rejection of messages at the processors, based upon destination data in the messages. All messages occupy places in a coherent priority scheme and are transferred in contending groups with prioritization on the network. Using data, status, control and response messages, and different multiprocessor modes, the system is particularly suited for configuration in a relational data base machine having capability for maintaining an extended data base and handling complex queries.

Claims

1. A multi-processor system having a plurality of processors and improved facility for intercommunication of data packets between processors, comprising:

a sort network, intercoupling the processors such that each of the processors is coupled to all others of the processors, for sorting the data packets into priority order based on at least one criteria comprising command code, data content, and data length;

means at each of the processors for concurrently presenting data packets to the sort network; and

means at each of the processors responsive to received data packets from the sort network for accepting packets based on the data content thereof.

2. A method of communicating between processors to confirm in a minimum number of steps that all processors intended to receive a given task message have correctly received the message, the method comprising the steps of:

concurrently transmitting competing task messages having a given order of priority from the processors;

sorting the task messages, while transferring them, to derive a priority task message delivered concurrently to all processors, wherein the sorting is based on at least one criteria comprising command code, data content, and data length;

concurrently responding to the priority task message at all processors with response messages having a given order of priority, a "correct reception" response message having a lower order of priority than a "rejection" response message for an intended processor, and the response messages having priority over the task messages;

sorting the response messages, while transferring them, to derive a single priority response message delivered to all intended processors, wherein the sorting is based on at least one criteria comprising command code, data content, and data length; and

accepting the priority task message at the intended processors if the priority response message is a "correct reception" response message.

3. The method as set forth in claim 2 above, wherein the "correct reception" response message comprises an "acknowledgement" response message and the "rejection" response message comprises a "non-acknowledgment" response message, and wherein processors for which the priority task message is not intended transmit a "not applicable" response message which is of lower priority than the "acknowledgment" response message.

4. The method as set forth in claim 3 above, wherein the "non-acknowledgment" response message indicates an Overrun, Transaction Number error or Lock condition.

5. The method as set forth in claim 4 above, further including the steps of locally checking a transaction number state to ascertain if a processor is in a "non-participant" or "initial" state, and sending a "not-applicable" response message otherwise.

6. The method as set forth in claim 2 above, wherein task messages which are not accepted are repeatedly transmitted and responded to until they are correctly received by all intended processors.

7. The method as set forth in claim 2 above, wherein the task messages are primary data messages and the response messages are locally generated by all processors and interspersed in other messages.

8. The method as set forth in claim 2 above, wherein the task messages terminate in "end of message" codes and the responses messages are transmitted by all processors concurrently after the "end of message" code.

9. A method for assuring, in a minimum number of interchanges, that all relevant processors in a multi-processor system accept or commit to a task, the method comprising the steps of:

broadcasting a message intended to be accepted by the relevant processors;

concurrently responding to the broadcast message with response messages indicating by data content thereof whether the broadcast message has been accepted at the relevant processors; and

arbitrating the response messages based on at least one criteria comprising command code, data content, and data length, to broadcast a single priority response message indicative of acceptance by all of the relevant processors, unless at least one relevant processor has indicated non-acceptance.

10. The method as set forth in claim 9 above, wherein each of the processors transmits response messages whose data contents vary to indicate a state of readiness of the processor, and wherein the arbitration step gives highest priority to a response message indicating a lowest state of readiness.

11. The method as set forth in claim 10 above, wherein each of the processors that is not relevant to the task responds with a low priority response message that loses in the arbitration step.

12. The method as set forth in claim 11 above, further including the step of generating, at the processors, concurrent competing messages which contain the tasks, and arbitrating among the competing messages to provide a single priority message for broadcast.

13. In a multi-processor system, a method of confirming successful concurrent acceptance of messages by different relevant processors to a given process in arbitrarily controlled groupings, comprising the steps of:

broadcasting a message to all of the processors;

recognizing, at each of the processors, the message'"'"'s relevance to the processor;

sending response messages concurrently from all processors, the response messages having data contents varying in accordance with a priority rule to denote that the broadcast message is not relevant, is relevant and reception is confirmed, or is relevant but reception is not confirmed; and

prioritizing the concurrent response messages based on at least one criteria comprising command code, data content, and data length, to establish whether all relevant processors have confirmed reception, such that one concurrent broadcast message to, and one concurrent response message from, each of the processors assures confirmation from all relevant processors.

14. A method of transmitting and receiving primary data messages in a multi-processor, the method comprising the steps of:

a) simultaneously broadcasting a primary data message to a plurality of processors;

b) determining at each individual processor if the broadcast primary data message is intended for the individual processor;

c) generating a response at each intended individual processor, the response indicating each individual processor'"'"'s ability to process the primary data message;

d) transmitting the generated response for each intended individual processor through a priority determining network to prioritize all the generated responses based on at least one criteria comprising command code, data content, and data length;

e) granting highest priority within the priority determining network to any generated response indicating any intended individual processor'"'"'s inability to process the primary data message; and

f) granting lowest priority within the priority determining network to any generated response acknowledging an intended individual processor'"'"'s ability to process the primary data message, thereby causing the priority determining network to create an output signal indicating acknowledgment of the primary data message only when intended individual processor indicates an ability to process the primary data message.

15. The method of transmitting and receiving primary data messages as set forth in claim 14, further comprising the step of detecting, at each individual processor, an "end of message" indicator residing within the primary data message.

16. The method of transmitting and receiving primary data messages as set forth in claim 15, wherein each individual processor initially determines if the primary data message is intended for the individual processor by identifying a destination select word residing within the primary data message.

17. The method of transmitting and receiving primary data messages as set forth in claim 16, wherein each individual processor finally determines if the primary data message is intended for the individual processor by identifying a transaction number residing within the primary data message.

18. A method of processing primary data messages in a multi-processor system, the method comprising the steps of:

a) transmitting the primary data messages to a network coupled to a plurality of processors;

b) sorting the primary data messages, by the network, into priority order based on at least one criteria comprising command code, data content, and data length to obtain a highest priority primary data message;

c) receiving the highest priority primary data message at the processors, the processors each independently;

i) determining when the reception of the highest priority primary data message is completed;

ii) determining if the processor is a member of a group of processors intended for the highest priority primary data message;

iii) determining if the intended processor is in a state to process the highest priority primary data message;

iv) generating a responsive message indicating if the intended processor will process the highest priority primary data message;

v) transmitting the responsive message to the network;

d) determining within the network if every intended processor will process the highest priority primary data message; and

e) completing the processing of the highest priority primary data message only if the network indicates that every intended processor will process the highest priority primary data message.

19. The method of claim 18, wherein the transmission of the primary data message to the processors occurs simultaneously.

20. The method of claim 19 above, wherein each of the processors determining when the primary data message is completed by identifying an "end of message" bit, the "end of message" bit residing within the primary data message.

21. The method of claim 20 above, wherein the intended processors will generate a responsive message that does not acknowledge the primary data message if the intended processors are occupied so as to prevent processing the primary data message.

22. The method of claim 21 above, wherein each of the processors determines if the processor is a member of the group of processors intended for the primary messages by performing the steps of:

a) identifying a destination selection word residing within the primary data message;

b) comparing the destination selection word to a destination map residing within a processor interface;

c) generating a responsive message indicating that the primary message is not applicable to the processor if the destination map does not contain the destination selection word; and

d) continuing to receive the primary data message if the destination map contains the destination selection word.

23. The method of claim 22 above, wherein each of the processors determines if the processor is in a state to process the primary data message by performing the steps of:

a) identifying a transaction number residing in a directory within the processor interface; and

b) determining if the transaction number corresponds to a "ready to process" state.

24. A method of processing tasks in different processors of a computer system and merging message packets into a sorted order, the method comprising the steps of:

transmitting messages to the different processors for identifying individual portions of the tasks to be performed;

asynchronously carrying out the individual portions of the task in the different processors to generate processed message packets;

locally chaining together the processed message packets at each of the different processors into a sorted order, thereby creating a plurality of sorted message packets;

transmitting a "start merge" instruction to the different processors concurrently;

concurrently transmitting a highest priority one of the sorted message packets from each of the different processors while dynamically prioritizing a single message packet based on at least one criteria comprising command code, data content, and data length; and

repeating the concurrent transmission and prioritizing with losing and new message packets until all message packets pertaining to the task have been transmitted in sorted order.

25. The method as set forth in claim 24 above, including the steps of identifying the task for the different processors by a transaction identity, and starting and stopping the merging of messages with control communications.

26. The method as set forth in claim 25 above, further including the step of using command messages to suspend and resume merge operations so that a number of copending merge operations can be performed simultaneously in the computer system.

27. The method as set forth in claim 26 above, further including the steps of determining the readiness of the different processors, passing control of the merge operations to a least ready of the processors, and the least ready of the processors ascertaining whether the merge operations can begin when it has completed its portion of the task.

28. The method as set forth in claim 27 above, further including the steps of transmitting "end of file" data from each of the different processors which has completed transmission of its message packets, and transmitting "end of merge" data when all of the different processors have sent "end of file" data.

29. The method as set forth in claim 28 above, further including the step of storing processed message packets together with next message vectors to designate the sequence of the locally chained processed message packets.

30. The method as set forth in claim 29 above, wherein a transaction identity other than zero establishes a merge mode of operation and a transaction identity of zero establishes a non-merge mode.

31. The method as set forth in claim 30 above, wherein the step of transmitting messages to the different processors comprises broadcasting the messages to all of the different processors in a non-merge mode and recognizing an applicability of the messages at each of the different processors.

32. The method as set forth in claim 31 above, wherein the computer system further comprises a data base and further including the step of updating the data base in accordance with the processed messages packets.

33. A method of placing messages from a plurality of individual processors in a sorted order, the messages being generated during the completion of asynchronous processing tasks, the method comprising the steps of:

assembling, at the individual processors, sequences of the messages, the messages having sorting criteria therein;

synchronously transmitting high priority messages according to the sorting criteria from each of the individual processors;

dynamically prioritizing competing high priority messages based on at least one criteria comprising command code, data content, and data length to reject all but a highest priority message; and

synchronously repeating the transmission of remaining high priority messages until all the messages have been completely transmitted.

34. The method as set forth in claim 33 above, further comprising the steps of starting and stopping the method by first transmitting start and stop commands of higher priority in competition with other messages, dynamically prioritizing the start and stop commands, and delivering the start and stop commands to all processors concurrently.

35. The method as set forth in claim 34 above, further comprising the steps of assembling messages in sorted sub-orders at the processors, dynamically prioritizing all messages pertaining to one sub-order, and then repeating the steps with subsequent sub-orders.

36. The method as set forth in claim 35 above, further comprising the steps of delivering the priority messages to all processors concurrently, responding to the priority message from all processors concurrently, and dynamically prioritizing the responses before synchronously transmitting further messages.

37. A method of multiplexing a multi-processing computer system so that the processors may function as to a member of task requests, the method comprising the steps of:

assigning each of the processors to a sort/merge stream for each task request with which the processor is to be active;

asynchronously assembling, at each of the processors, sorted subsets of messages pertaining to the relevant task requests for that processor;

synchronously transmitting high priority messages from each subset pertaining to a given task request;

dynamically sort/merging the competing high priority messages to select a highest priority message wherein the high priority messages are sorted based on at least one criteria comprising command code data content and data length; and

assembling the sort/merge stream by repeated synchronous transmissions and selection of successive high priority messages.

38. The method as set forth in claim 37 above, further comprising the steps of changing the assembly of sort/merge streams by transmitting start and stop commands of a higher priority relating to different task requests, dynamically sort/merging the commands and broadcasting the commands to all of the processors simultaneously.

39. The method as set forth in claim 38 above, including the steps of identifying task requests with transaction numbers, using a specific transaction number as a non-merge command, and transmitting"start merge" commands, so that global merge operations and broadcast transmissions can be intermingled as to different task requests to enable global coordination of system resources.