Task scheduler for a fault tolerant multiple node processing system

• US 4,805,107 A
• Filed: 04/15/1987
• Issued: 02/14/1989
• Est. Priority Date: 04/15/1987
• Status: Expired due to Fees

• Current Assignee: Allied Corp. (Honeywell International Inc.)
• Original Assignee: Alliedsignal Inc.
• Inventors: Finn, Alan M., Walter, Chris J., Kieckhafer, Roger M.
• Primary Examiner(s): NOT, DEFINED
• Application Number:
  US07/039,190
• Time in Patent Office: 671 Days
• Field of Search: 364/200, 364/900
• US Class Current: 714/4.3
• CPC Class Codes: G06F 11/0724 in a multiprocessor or a multi-core unit multiprocessors per se G06F15/80
• CPC Class Codes: G06F 11/076 by exceeding a count or rate limit, e.g. word- or bit count limit
• CPC Class Codes: G06F 11/10 Adding special bits or symbols to the coded information, e.g. parity check, casting out 9's or 11's
• CPC Class Codes: G06F 11/1425 by reconfiguration of node membership
• CPC Class Codes: G06F 11/1482 by means of middleware or OS functionality
• CPC Class Codes: G06F 11/1658 Data re-synchronization of a redundant component, or initial sync of replacement, additional or spare unit
• CPC Class Codes: G06F 11/181 Eliminating the failing redundant component
• CPC Class Codes: G06F 11/182 based on mutual exchange of the output between redundant processing components
• CPC Class Codes: G06F 11/187 Voting techniques
• CPC Class Codes: G06F 11/188 where exact match is not required
• 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 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 9/4881 Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues

Abstract

A task scheduler for a fault tolerant multiple node processing system having a task activity list storing a set of application tasks, a priority scan list storing a selected portion of the set of application tasks, a completion status list also storing the same selected portion of the set of application tasks. A wake-up sequencer transfers the application tasks from the task activity list to the priority scan list, and a priority scanner transfers the application tasks ready for execution from the priority scan list to a selection queue. A next task selector selects the next application task that its node will execute, and a task started register stores the identity of the application tasks completed by the other nodes. A task interactive consistency (TIC) handler updates the status of the application tasks stored in the task activity list, the priority scan list, and the completion status list in response to messages received from the other nodes identifying which nodes completed tasks. The task interactive consistency handler checks the scheduling process of each node by comparing the application tasks it reported to have started with the highest priority application task scheduled for that node in its selection queue.

Claims

1. In a mutliple node fault tolerant processing system for processing a set of application tasks in which each node has an applications processor for executing a predetermined subset of said set of application tasks and an operations controller for controlling its own node and scheduling the application tasks in said predetermined subset of tasks that are to be executed by the applications processor through an exchange of inter-node messages containing data and operation information with each node in the processing system, the operations controller generating at least two timing period intervals, a fundamental timing period and a master period which is an integer multiple of the fundamental timing period, the master period defining a timing interval during which every task in said predetermined subset of tasks is scheduled for execution by the applications processor at least once, each operations controller having a task scheduler comprising:

a task activity list containing an entry for each application task in said multiple node processing system, each entry containing an execution periodicity and a node allocation for that application task;

a priority scan list containing a selected portion of said application tasks in the task activity list which are available for execution, said selected portion of said application tasks being stored in their preferred order of execution;

a completion status list storing said selected portion of said application tasks stored in said priority scan list;

a selection queue storing for each node said application tasks ready for selection in their preferred order of execution;

a period counter for counting said fundamental timing periods to generate a period count corresponding to a number of fundamental periods which have expired since a beginning of a new master period;

wake-up sequencer means connected to said task activity list, said priority scan list, said completion status list and said period counter for interrogating said task activity list to transfer to said priority scan list and said completion status list all of said application tasks whose periodicity is greater than said period count;

priority scan means connected between said priority scan list and said selection queue for transferring to said selection queue for each node entry three application tasks which are ready for execution by that node, which have a highest priority in said preferred order of execution;

task selector means connected to said selection queue for selecting in said preferred order of execution, said application task currently stored in said selection queue for its own node as a next task scheduled for execution by its own applications processor; and

a task interactive consistency handler connected to said task activity list, said priority scan list, said completion status list and said selection queue for updating the status of each task in said task activity list, said priority scan list, said completion status list and said selection queue which are identified in inter-node messages reporting the completion of a task.

2. The task scheduler of claim 1 wherein each application task entry of said completion status list has a completion count entry storing a 2'"'"'s complement of a number of nodes which are scheduled to execute that application task, said task interactive consistency handler having means for incrementing said completion count in response to inter-node messages identifying which node completed that application task and for setting a terminated flag when said completion count is decremented to zero indicating that the task has been executed by each of the nodes scheduled to execute that task.

3. The task scheduler of claim 1 wherein each task entry of said task activity list and said priority scan list has a predecessor count entry indicative of a number of tasks which must be terminated before it can be executed, said task interactive consistency handler having a successor list storing an identity of all the tasks for which a terminated task is a predecessor, and means responsive to a termination of a task for accessing said successor list to identify each task for which said terminated task is a predecessor, and for decrementing the predecessor count in said task activity list and said priority scan list for each of said identified tasks.

4. The task scheduler of claim 3 further having an "old task" table connected to said task interactive consistency handler, said "old task" storing for each node the task currently being executed by that node, said task interactive consistency handler having means for recording as "used" in said selection queue a task having a highest priority in said preferred order of execution currently stored for each node which reported it has started a new task and for recording an identity of said task having said highest priority in said "old task" table.

5. The task scheduler of claim 1 in which the inter-node messages exchanged between said multiple nodes includes a task completed/started message and a task interactive consistency message, said task completed/started message is sent to each node in said multiple node processing system whenever a node begins a new task, said task completed/started message containing at least an identity of the task started and an identity of the task completed by that node and said task interactive consistency message is sent at predetermined timing intervals and contains a task completed vector identifying each node which sent a task completed/started message, said task completed vector being a voted composite of said task completed/started messages received from each of said multiple nodes in a predetermined timing interval, said task scheduler having a started task register connected to said task interactive consistency handler for storing said identity of the task reported as started in said task completed/started messages received from that node and said task interactive consistency handler responsive to said task completed vector contained in said task interactive consistency message to compare an identity of the task having said highest priority stored for each node identified as having completed a task in said task completed vector with an identity of the task stored in said task started register and to generate a sequence error signal said identity of the task having said highest priority is different from said identity of the task in said task completed vector.

6. The task scheduler of claim 5 having a Byzantine voter voting on the task completed vectors received in the task interactive consistency messages from all the nodes to generate a Byzantine voted task completed vector.

7. The task scheduler of claim 6 having an "old task" table connected to said task interactive consistency handler storing for each node the task currently being executed by that node, said task interactive consistency handler having means for recording as "used" in said selection queue the task having said highest priority currently stored for each node identified in said Byzantine voted task completed vector as having completed a task and for recording in said "old task" table for each node identified in said Byzantine voted task completed vector, an identity of the task just recorded as "used" in said selection queue.

8. The task scheduler of claim 7 wherein each task entry of said completion status list has a completion count entry storing a 2'"'"'s complement of a number of nodes which are scheduled to execute that task, said task interactive consistency handler accessing said "old task" table to obtain an identity of a current task for each node identified as having completed a task by said Byzantine voted task completed vector and using said identity of said current task to access said completion status list to increment said completion count entry, said task interactive consistency handler including means for setting a terminated flag when said completion count entry is decremented to zero and for generating an internal message identifying each task whose terminated flag is set.

9. The task scheduler of claim 8 wherein each task of said set of application tasks has an associated maximum execution time and an associated minimum execution time and wherein said task scheduler has a plurality of execution timers connected to said "old task" table and said interactive consistency handler, one execution timer for each node for storing a 2'"'"'s complement of the execution time for a current task, said task interactive consistency handler being responsive to said Byzantine voted task completed vector to generate an execution time error for each node identified as having completed a task whose associated maximum execution time is greater than said execution time stored in its associated execution timer, said task interactive consistency handler further loading said associated execution timer with a 2'"'"'s complement of the maximum execution time for a next highest priority unused task in the selection queue for that node, each of said execution timers being periodically incremented and operative to generate a timing error signal when said execution time is incremented to zero.

10. The task scheduler of claim 9 wherein each task entry of said task activity list and said priority scan list includes a predecessor count corresponding to a number of predecessor tasks that must be completed before that task can be executed, and an allocation vector identifying each node on which that task can be executed, said task activity list further having a periodicity number corresponding to a number of fundamental timing periods which must pass before the task can be executed again, and said priority scan list having an iteration entry corresponding to a number of times that task has been executed, said wake-up sequencer comparing said period count generated by said period counter with the periodicity stored for each task to transfer said predecessor count and said allocation vector from said task activity list to an appropriate entry in said priority scan list.

11. The task scheduler of claim 10 wherein said selection queue stores three entries for each node in the processing system, each entry having a "used" flag, an iteration count and a task identification code, said priority scan means scans said priority scan list to transfer to said selection queue, for each node, said three tasks having said highest priority whose precedent count is zero and whose allocation vector indicates the task is to be executed by that node, said tasks being stored in said selection queue in said order of preferred execution.

12. The task scheduler of claim 11 wherein said selection queue has three pages, a NEXT page from which said task selector means selects the tasks scheduled for execution by the applications processor, a PREVIOUS page, and a CHECK page, each of said three pages containing three entries for each node, and each entry having a "used" bit which is set to true when said task selector means selects that task, an iteration number which specifically identifies this task from prior executions of said task, and a task identification code which uniquely identifies that task from all other tasks.

13. The task scheduler of claim 12 wherein said task interactive consistency handler marks as "used" on said CHECK page, an unused task having said highest priority for each node identified in said Byzantine voted task completed vector as having completed a task and records the task identification code of the entry just marked "used" in the entry for that same node in said "old task" table, said task interactive consistency handler further marking as "used" on said PREVIOUS and NEXT pages for each node said task interactive consistency handler marked "used" on said CHECK page.

14. The task scheduler of claim 13 wherein said operations controller has a sub-atomic period timing interval which is a subdivision of said fundamental timing period, said task interactive consistency handler transferring each entry of said PREVIOUS page to said CHECK page and transferring each entry of said NEXT page to said PREVIOUS page at a beginning of each of said subatomic period timing intervals prior to said priority scan means transferring a next set of highest priority tasks to said NEXT page for each node.

15. The task scheduler of claim 13 wherein said operations controller has a sub-atomic period timing interval which is a subdivision of said fundamental timing period, and said task interactive consistency handler has a plurality of pointers, one for each of said NEXT, PREVIOUS and CHECK pages of said selection queue, said task interactive consistency handler rotating said plurality of pointers such that the NEXT page becomes the PREVIOUS page, the PREVIOUS page becomes the CHECK page, and the CHECK page becomes the NEXT page at a beginning of each sub-atomic period prior to said priority scan means transferring a next set of highest priority tasks to said NEXT page for each node.

16. The task scheduler of claim 13 wherein said task interactive consistency message further includes a branch condition bit for each node which completed a task, said branch bit being a 1-bit identifying a first of two possible successors for the completed task and a 0-bit identifying a second of said two possible successors, said Byzantine voter also performs a Byzantine vote on said branch condition bits to generate a Byzantine voted branch byte containing voted values of said branch condition bits for each node, and wherein each entry of said completion status list has a completion count entry storing a number of nodes which completed the task, a branch count entry storing a number of nodes whose branch condition bit for that task is a 1-bit and an allocation vector, said task interactive consistency handler obtaining for each node identified in said Byzantine voted task completed vector as having completed a task, an identity of a completed task from said "old task" table, and updating the status of said completed tasks in said completion status list by incrementing the completion count once for each node that completed that task, by adding said Byzantine voted branch bit to said branch count and by setting to "false" an allocation bit in said allocation vector.

17. The task scheduler of claim 16 wherein said successor list has for each completed task a first successor task list containing the successor tasks to be executed when said branch condition bit is a 1-bit and a second successor task list containing the successor tasks to be executed when said branch condition bit is a 0-bit, said task interactive consistency handler comparing said branch count with one-half the value of said completion count, in response to all the allocation flags being set to "false," to determine which branch condition bit is agreed on by a majority of nodes which completed the task, accessing said successor task list for that task and using the branch condition bit of said majority to get an identification code for a successor task and using said task identification code to access said successor task in said task activity list and said priority scan list to increment the predecessor count for that task.

18. The task scheduler of claim 1 wherein said operations contoller has means for excluding from and readmitting nodes to active participation in said multiple node processing system and said operations controller further has means for periodically generating a system state vector identifying each node currently excluded from participating in said multiple node processing system, said task scheduler further having a reconfiguration module comprising:

a system state comparator for comparing each received system state vector with an immediately preceding system state vector to generate a delta vector identifying each newly excluded or readmitted node;

a relevance table storing a relevance vector for each task identifying which nodes may execute the task;

a swap table storing a predecessor count, said periodicity and a swap count for each task, said swap count corresponding to a number of nodes relative to that task that may be excluded or readmitted before its swap status is to be complemented;

a task swapper connected to said swap table and said task activity list for identifying each task to which the node identified in the delta vector was relevant to test said swap count and to complement its swap status if said swap count is zero, said swap status being indicative of whether the task will be included in said task activity list or removed from said task activity list in the system state identified by the system state vector, said task swapper further decrementing said swap count in response to the node identified in the delta vector being excluded and incrementing said swap count in response to the node identified in said delta vector being readmitted;

an allocation table storing an allocation count for each node-task combination indicative of the current node allocation status of each task;

a preferred table storing a preferred vector for each task listing the nodes preferred for the execution of that task;

task reallocator means connected to said allocation table, said preferred table and said task activity list for determining for each node, from said preferred table, if the node identified in the delta vector was more preferred for the execution of that task than itself, for complementing the allocation status for that node when the node is more preferred than the node identified in the delta vector and said allocation count is zero, for decrementing said allocation count when said node identified in said delta vector is more preferred and is an excluded node and for incrementing said allocation count when said node identified in said delta vector is more preferred and is a readmitted node; and

means for recording in said task activity list said predecessor count and the periodicity from the swap table and the allocation status of each task from said allocation table whose swap status indicates that the task should be active in the next system state of the system.

19. The task scheduler of claim 18 wherein said set of applications tasks includes a first set of tasks to be executed only by said non-excluded nodes and a second set of tasks to be executed only by the excluded nodes and said relevance table includes an excluded flag bit which identifies each task in said second set of tasks, said reconfiguration module further includes a task status matcher which will compare for each task, the status of each node identified in said next system state vector with said excluded flag bit to generate a match status bit enabling said means for recording to record that task in said task activity list.