Method and apparatus for analyzing error conditions in a massively parallel computer system by identifying anomalous nodes within a communicator set

• US 7,930,595 B2
• Filed: 06/22/2006
• Issued: 04/19/2011
• Est. Priority Date: 06/22/2006
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Gooding, Thomas Michael
• Primary Examiner(s): Baderman; Scott T
• Assistant Examiner(s): Leibovich; Yair
• Application Number:
  US11/425,773
• Publication Number:
  US 20080022261A1
• Time in Patent Office: 1,762 Days
• Field of Search: 714/10, 714/37
• US Class Current: 714/37
• CPC Class Codes: G06F 11/36 Preventing errors by testing or debugging software

Abstract

An analytical mechanism for a massively parallel computer system automatically analyzes data retrieved from the system, and identifies nodes which exhibit anomalous behavior in comparison to their immediate neighbors. Preferably, anomalous behavior is determined by comparing call-return stack tracebacks for each node, grouping like nodes together, and identifying neighboring nodes which do not themselves belong to the group. A node, not itself in the group, having a large number of neighbors in the group, is a likely locality of error. The analyzer preferably presents this information to the user by sorting the neighbors according to number of adjoining members of the group.

Claims

1. A computer-implemented method for analyzing errors in a parallel computer system, said parallel computer system comprising multiple nodes arranged in a lattice for inter-nodal communications, each node comprising at least one processor for executing a respective application sub-process and a nodal memory, said method comprising the steps of:

executing a respective unique application sub-process of a common application in each node of a plurality of said nodes of said parallel computer system to produce respective independent state data in each node of said plurality of said nodes;

obtaining said respective independent state data corresponding to each node of said plurality of nodes of said parallel computer system;

analyzing said independent state data to identify a first node having anomalous corresponding independent state data with respect to respective independent state data corresponding to a plurality of neighboring nodes of said first node, each said neighboring node being a node adjacent said first node within said lattice, wherein said step of analyzing said independent state data to identify a first node comprises;

identifying a first subset of said plurality of nodes, said first subset consisting of all nodes having independent state data which matches according to a pre-defined matching criterion; and

using the topology of said lattice to identify said first node as an anomalous neighbor of at least one node of said first subset; and

presenting results of said analyzing step to a user.

2. The method for analyzing errors of claim 1, wherein said respective independent state data comprises a respective call-return stack traceback for each of said plurality of nodes.

3. The method for analyzing errors of claim 2, wherein said step of analyzing said independent state data comprises identifying a plurality of discrete subsets of said plurality of nodes for which corresponding respective independent state data was obtained by said step of obtaining respective independent state data, each subset comprising one or more nodes having matching call-return stack tracebacks according to said pre-defined matching criterion.

4. The method for analyzing errors of claim 3, wherein said step of analyzing said independent state data further comprises determining, with respect to each respective node of a plurality of nodes, a number of neighbors of the respective node which are members of said first subset, said first subset being one of said plurality of discrete subsets;

and

identifying said first node according to said number of neighbors and membership in said first subset.

5. The method for analyzing errors of claim 1, wherein said step of using the topology of said lattice to identify said first node as an anomalous neighbor of at least one node of said first subset comprises:

determining, with respect to each of a plurality of nodes, a number of neighbors of the respective node which are members of said first subset; and

identifying said first node according to said number of neighbors.

6. The method for analyzing errors of claim 1, wherein said step of analyzing said independent state data to identify a first node comprises:

for each respective node of a second subset of said plurality of nodes, identifying a number of neighbors of the respective node which are members of said first subset of said plurality of nodes.

7. The method for analyzing errors of claim 6,

wherein said step of analyzing said independent state data to identify a first node further comprises sorting said nodes of said second subset according to the respective number of neighbors of each node of said second subset which are members of said first subset;

and

wherein said step of presenting results of said analyzing step to a user comprises presenting said second subset of nodes in order of said sorting.

8. A program product for analyzing errors in a parallel computer system, said parallel computer system comprising multiple nodes arranged in a lattice for inter-nodal communications, each node comprising at least one processor for executing a respective application sub-process and a nodal memory, the program product comprising:

a plurality of computer executable instructions recorded on tangible computer-readable storage media, wherein said instructions, when executed by at least one computer system, cause the at least one computer system to perform the steps of;

receiving respective independent state data corresponding to each of a plurality of said nodes, said respective independent state data being produced as a result of executing a respective unique application sub-process of a common application in each node of said plurality of nodes of said parallel computing lattice;

analyzing said independent state data to identify a first node of said plurality of said nodes having anomalous corresponding independent state data with respect to respective independent state data corresponding to a plurality of neighboring nodes of said first node, each said neighboring node being a node adjacent said first node within said lattice, wherein said analyzing said independent state data to identify a first node comprises;

identifying a first subset of said plurality of nodes using said state data, said first subset consisting of all nodes having respective independent state data which matches according to a pre-defined matching criterion; and

using the topology of said lattice to identify said first node as an anomalous neighbor of at least one node of said first subset; and

presenting results of said analyzing step to a user.

9. The computer program product of claim 8, wherein said respective independent state data comprises a respective call-return stack traceback for each of said plurality of nodes.

10. The computer program product of claim 8,

wherein said step of analyzing said independent state data to identify a first node of said plurality of nodes comprises sorting said nodes of said plurality of nodes according to the respective number of neighbors of each node which are members of said first subset;

and

wherein said step of presenting results of said analyzing step to a user comprises presenting said first nodes for which a number of neighbors was determined in order of said sorting.

11. A computer system which analyzes errors in a parallel computing lattice, said lattice comprising a plurality of nodes coupled by inter-nodal communications paths, each node comprising at least one processor for executing a respective application sub-process and a nodal memory, the computer system comprising:

at least one processor;

a memory for storing data addressable by said at least one processor;

an analytical program embodied as computer executable instructions storable in said memory and executable on said at least one processor, said analytical program comprising;

(a) a state data function which receives respective independent state data corresponding to each of a plurality of nodes of said parallel computing lattice, said respective independent state data being produced as a result of executing a respective unique application sub-process of a common application in each node of said plurality of nodes of said parallel computing lattice;

(b) an anomaly detector function which identifies a first node having anomalous corresponding independent state data with respect to respective independent state data corresponding to a plurality of neighboring nodes of said first node, each said neighboring node being a node adjacent said first node within said lattice, wherein said anomaly detector function identifies a first subset of said plurality of nodes, said first subset consisting of all nodes having state data which matches according to a pre-defined matching criterion, and wherein said anomaly detector function uses the topology of said lattice to identify said first node as an anomalous neighbor of at least one node of said first subset; and

(c) an output function which presents results of said state data and anomaly detector functions to a user.

12. The computer system of claim 11, wherein said computer system is a massively parallel computer system which includes said parallel computing lattice.

13. The computer system of claim 12, wherein said computer system comprises a control subsystem separate from said parallel computing lattice, said control subsystem including at least one processor, a memory for storing data addressable by said at least one processor, and said analytical program.

14. The computer system of claim 11, wherein said respective independent state data comprises a respective call-return stack traceback for each of said plurality of nodes.

15. The computer system of claim 14, wherein said anomaly detector function identifies a plurality of discrete subsets of said plurality of nodes for which corresponding respective independent state data was received by said state data function, each subset comprising one or more nodes having matching call-return stack tracebacks according to said pre-defined matching criterion.

16. The computer system of claim 11, wherein said anomaly detector function identifies said first node as an anomalous node by determining, with respect to each of a plurality of nodes, a number of neighbors of the respective node which are members of said first subset, and identifying said first node according to said number of neighbors.

17. The computer system of claim 11, wherein said anomaly detector function determines, for each respective node of a second subset of said plurality of nodes, a number of neighbors of the respective node which are members of said first subset of said plurality of nodes.