TEMPLATE BASED PARALLEL CHECKPOINTING IN A MASSIVELY PARALLEL COMPUTER SYSTEM

• US 20080215916A1
• Filed: 04/16/2008
• Published: 09/04/2008
• Est. Priority Date: 04/14/2005
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Archer, Charles Jens, Inglett, Todd Alan
• Granted Patent: US 7,627,783 B2
• Time in Patent Office: Days
• US Class Current: 714/15
• CPC Class Codes: G06F 11/1438 Restarting or rejuvenating
• CPC Class Codes: G06F 11/1451 by selection of backup contents

Abstract

A method and apparatus for a template based parallel checkpoint save for a massively parallel super computer system using a parallel variation of the rsync protocol, and network broadcast. In preferred embodiments, the checkpoint data for each node is compared to a template checkpoint file that resides in the storage and that was previously produced. Embodiments herein greatly decrease the amount of data that must be transmitted and stored for faster checkpointing and increased efficiency of the computer system. Embodiments are directed to a parallel computer system with nodes arranged in a cluster with a high speed interconnect that can perform broadcast communication. The checkpoint contains a set of actual small data blocks with their corresponding checksums from all nodes in the system. The data blocks may be compressed using conventional non-lossy data compression algorithms to further reduce the overall checkpoint size.

Claims

1. ) A massively parallel computer system comprising:

a) a plurality of compute nodes;

b) a plurality of input/output (I/O) nodes connected to the compute nodes;

c) a network that connects the compute nodes and the I/O nodes that supports a broadcast communication with the compute nodes; and

d) a checkpoint server that collects a parallel checkpoint of the state of the compute nodes using a rolling checksum algorithm and broadcasts a list of data block checksums for a plurality of data blocks on the server from a previous checkpoint to the plurality of compute nodes and the compute nodes search their own memory image for checksum matches using the rolling checksum algorithm.

2. ) The massively parallel computer system of claim 1 wherein the compute nodes search their own memory image for checksum matches using an rsync protocol rolling checksum algorithm.

3. ) The massively parallel computer system of claim 2 wherein each compute node produces a template of new data blocks with checksums that didn'"'"'t exist in the previous checkpoint and a template of references to the original data blocks that did exist in the previous checkpoint where the template on the compute node identifies which of the memory blocks stored in the checkpoint server make up the memory of the compute node.

4. ) The massively parallel computer system of claim 3 wherein each node sends its new data block checksum template to an adjacent node in the cluster of nodes;

and compares checksums to find common data blocks between all adjacent nodes as well as its own data blocks; and

informs adjacent nodes to replace a reference to a common data block with a reference to a data block on another node.

5. ) The massively parallel computer system of claim 4 wherein each node collects reference templates from the compute nodes and stores them in the checkpoint server;

and

collects new unique data blocks and stores them to the checkpoint server.

6. ) The massively parallel computer system claim 2 wherein the I/O nodes store the data blocks to unique storage servers over network lines connected to the I/O nodes.

7. ) A program product comprising:

(A) a checkpoint server that broadcasts a list of data block checksums for a plurality of data blocks on a server from a previous checkpoint to all compute nodes in a cluster and the compute nodes search their own memorv image for checksum matches and collects a parallel checkpoint of the state of a cluster of compute nodes on a massively parallel computer system using a rolling checksum algorithm on each node; and

(B) computer recordable media bearing the checkpoint server.

10. ) The program product of claim 7 wherein the compute nodes search their own memory image for checksum matches using an rsync protocol rolling checksum algorithm.

11. ) The program product of claim 10 wherein each compute node produces a template of new data blocks with checksums that didn'"'"'t exist in the previous checkpoint and a template of references to the original data blocks that did exist in the previous checkpoint where the template on the compute node identifies which of the memorv blocks stored in the checkpoint server make up the memorv of the compute node.

12. ) The program product of claim 11 wherein each node sends its new data block checksum template to an adjacent node in the cluster of nodes;

and compares checksums to find common data blocks between all adjacent nodes as well as its own data blocks; and

informs adjacent nodes to replace a reference to a common data block with a reference to a data block on another node.

13. ) The program product of claim 12 wherein each node collects reference templates from the compute nodes and stores them in the checkpoint server;

and collects new unique data blocks and stores them to the checkpoint server.

14. ) The program product of claim 13 wherein the I/O nodes store the data blocks to unique storage servers over network lines connected to the I/O nodes.

8. ) (canceled)

9. ) (canceled)