Process stop method and apparatus for a distributed memory multi-processor system

A process stop method for a distributed processing system that has a plurality of nodes that are interconnected with each other in a network, wherein said nodes have a particular ranking with respect to each other, the method comprising the steps of:

• a) transmitting data from one of said nodes to another of said nodes to perform a parallel processing task based on a thread involving at least two of said nodes;
• b) determining, by said one of said nodes, if said another of said nodes has acknowledged receipt of the data transmitted in the step a);
• d) if said another of said nodes has not acknowledged receipt of the data within a fixed time period, checking to determine if said another node has been stopped by an externally-received stop request to said another node, wherein said checking is only performed if said another node has a higher ranking than said one node;
• and
• e) if said checking results in said another node having been stopped, stopping said one node and informing the stopping of said one node to said intranodal synchronizing unit to be used by others of said at least two of said nodes involved in the parallel processing task,
• wherein, if said another node has not acknowledged receipt of the data within a fixed time period and if said another node does not have a higher ranking than said one node, determining that said another node has not received said data due to something other than a stop request sent to said another node, and returning to step a) to retransmit the data to said another node.

A process stop method for a distributed processing system that has a plurality of nodes that are interconnected with each other in a network, wherein said nodes have a particular ranking with respect to each other, the method comprising the steps of:

• a) receiving, by one of said nodes, a stop request for stopping a thread involving at least another node, said stop request being provided from an external source with respect to the processing system;
• b) based on said stop request, determining whether any current data communication are being performed by said one node to said at least another node;
• c) if the determination in step b) is that no other current data communications are being performed, stopping operation at said one node with respect to said thread;
• d) if the determination in step b) is that other current data communications are being performed by said one node, sending an indicator to an intranodal synchronizing unit accessible by at least said one node and said at least another node in said processing system, said indicator being used by said at least another node to determine that said one node will not at present participate in any communications with said at least another node due to said stop request resulting in stoppage at said one node.

A process stop apparatus for a distributed memory multi-processor system that includes a plurality of nodes interconnected into a network, each of said nodes being provided with a thread for parallel processing, wherein each of said nodes including a management process for managing a thread of each of said nodes, said management process for each of said nodes comprising:

• intranodal stop requesting means for requesting a stop from said thread of each of said nodes in response to a stop request received by each of said nodes;
• intranodal stop waiting means for waiting until every thread within each of said nodes makes a stop;
• and
• internodal stop requesting means for request a stop from the management process of a lower ranking node.

At least one recording medium for storing a program causing a plurality of nodes interconnected within a network to execute a process stop processing, each of said nodes being provided with one or more threads for parallel processing, wherein each of said nodes further being provided with a management process for managing said one or more threads of said each of said nodes, said program comprising the steps of:

• requesting a stop from said thread of each of said nodes in response to a stop request received by one of said nodes;
• waiting until an indication has been received that every thread within said one of said nodes has made a stop;
• and
• outputting a request for a stop to the management process of a node that is a next lower ranking node with respect to said one of said nodes.

At least one recording medium for storing a program causing a plurality of nodes interconnected within a network to execute a process stop processing, each of said nodes being provided with one or more threads for parallel processing, wherein each of said nodes further being provided with a management process for managing said one or more threads of said each of said nodes, said program comprising the steps of:

• receiving, by one of said nodes, a stop request for a particular thread;
• providing a checkpoint requesting flag to said particular thread based on said stop request;
• waiting until an indication has been received that said particular thread has made a stop;
• after said indication has been received that said particular thread has made the stop, informing a next lower ranking node of the stop request;
• making a determining that all of said nodes involved in said particular thread that have a lower ranking than said one of said nodes have made a stop, the determination being made based on the stop request being relayed to said all of said nodes involved in said particular thread;
• and
• after determining that said all of said nodes involved in said particular thread that are lower ranking than said one of said nodes have made the stop, stopping said one node and informing a next higher ranking node involved in said particular thread of the stop of said one node.

At least one recording medium for storing a program causing a plurality of nodes interconnected within a network to execute a process stop processing, each of said nodes being provided with one or more threads for parallel processing, wherein each of said nodes further being provided with a management process for managing said one or more threads of said each of said nodes, said program comprising the steps of:

• receiving a checkpoint request in the parallel processing;
• based on the checkpoint request, requesting a stop from a thread of a highest ranking node of said plurality of nodes;
• confirming that the thread of the highest ranking node has stopped;
• delivering a stop request to a next highest ranking node with respect to the highest ranking node;
• and
• determining when all of said nodes that are involved in the parallel processing have made a stop, so as to achieve an orderly process stop of the parallel processing.