Method, apparatus, and computer program product for design and selection of an I/O subsystem of a supercomputer

• US 9,652,568 B1
• Filed: 11/13/2012
• Issued: 05/16/2017
• Est. Priority Date: 11/14/2011
• Status: Active Grant

• Current Assignee: Emc IP Holding Company LLC (Dell Technologies Inc.)
• Original Assignee: Emc IP Holding Company LLC (Dell Technologies Inc.)
• Inventors: Faibish, Sorin, Gupta, Uday K., Tzelnic, Percy, Bent, John
• Primary Examiner(s): Phan, Thai
• Assistant Examiner(s): Khan, Iftekhar
• Application Number:
  US13/675,965
• Time in Patent Office: 1,645 Days
• CPC Class Codes: G06F 15/177 Initialisation or configuration control processor initialisation G06F9/4405
• CPC Class Codes: G06F 16/00 Information retrieval; Database structures therefor; File system structures therefor
• CPC Class Codes: G06F 2111/20 Configuration CAD, e.g. designing by assembling or positioning modules selected from libraries of predesigned modules
• CPC Class Codes: G06F 30/33 Design verification, e.g. functional simulation or model checking

Abstract

For simulating a parallel supercomputing cluster, a simulation program includes a model of the system of compute nodes and a model of an I/O subsystem that stores checkpoints from the compute nodes. When executed, instructions in the simulation program perform the steps of receiving input parameters defining the compute nodes and the I/O subsystem, computing a total number of computational flops for a time between checkpoints and an amount of disk storage required to store the checkpoint data for different configurations of the parallel supercomputing cluster, and presenting a summary to a user of the computed number of computational flops for a time between checkpoints and an amount of disk storage required to store the checkpoint data for the different configurations of the parallel supercomputing cluster.

Claims

1. An apparatus for simulating a parallel supercomputing cluster having a system of compute nodes and an I/O subsystem connected to the compute nodes for storing checkpoint data from the compute nodes, the apparatus comprising a digital computer including a data processor and non-transitory computer readable storage medium storing a simulation program, the simulation program including a model of the system of compute nodes and a model of the I/O subsystem and also including computer instructions that, when executed by the data processor, perform the steps of:

(a) receiving input parameters defining the system of compute nodes and input parameters defining the I/O subsystem;

(b) computing a total number of computational floating point operations for a time between checkpoints for different configurations of the parallel supercomputing cluster, and computing an amount of disk storage required to store the checkpoint data for the different configurations of the parallel supercomputing cluster; and

(c) presenting, to a user, a summary of the computed total number of computational floating point operations for the time between checkpoints for the different configurations of the parallel supercomputing cluster, and presenting, to the user, a summary of the computed amount of disk storage required to store the checkpoint data for the different configurations of the parallel supercomputing cluster.

2. The apparatus as claimed in claim 1, wherein the input parameters defining the system of compute nodes include a specified checkpoint target time defining an interval of time between checkpoints.

3. The apparatus as claimed in claim 1, wherein the different configurations of the parallel supercomputing cluster include a configuration using Lustre file system clients, running in the compute nodes.

4. The apparatus as claimed in claim 1, wherein the different configurations of the parallel supercomputing cluster include a configuration using Parallel Log-Structured File System (PLFS) on the compute nodes on top of Lustre file system mounted on each compute node.

5. The apparatus as claimed in claim 1, wherein the different configurations of the parallel supercomputing cluster include a configuration of the I/O subsystem using Parallel Log-Structured File System (PLFS) on the compute nodes and writing to a burst buffer that each compute node is connected to.

6. The apparatus as claimed in claim 5, wherein the model of the I/O subsystem has checkpoint data written to the burst buffer using Message Passing Interface Input-Output (MPI-IO) during an interval of time between checkpoints when an MPI job is paused, and the burst buffer is drained to a Lustre file system during the MPI job compute time during the interval of time between checkpoints.

8. The apparatus as claimed in claim 1, wherein the summary of the computed total number of computational floating point operations presented to the user includes graphics of a time series of floating point operations per second and a total number of floating point operations.

7. An apparatus for simulating a parallel supercomputing cluster having a system of compute nodes and an I/O subsystem connected to the compute nodes for storing checkpoint data from the compute nodes, the apparatus comprising a digital computer including a data processor and non-transitory computer readable storage medium storing a simulation program, the simulation program including a model of the s stem of compute nodes and a model of the I/O subs stem and also including computer instructions that, when executed by the data processor, perform the steps of:

(a) receiving input parameters defining the s stem of compute nodes and input parameters defining the I/O subsystem;

(b) computing a total number of computational floating point operations for a time between checkpoints for different configurations of the parallel supercomputing cluster, and computing an amount of disk storage required to store the checkpoint data for the different configurations of the parallel supercomputing cluster; and

(c) presenting, to a user, a summary of the computed total number of computational floating point operations for the time between checkpoints for the different configurations of the parallel supercomputing cluster, and presenting, to the user, a summary of the computed amount of disk storage required to store the checkpoint data for the different configurations of the parallel supercomputing cluster;

wherein step (b) includes computing an amount of time required to write checkpoint data from the compute nodes to the I/O subsystem for each of the different configurations of the parallel supercomputing cluster, and building a time series for each of the different configurations of the parallel supercomputing cluster.

9. A method of simulating a parallel supercomputing cluster having a system of compute nodes and an I/O subsystem connected to the compute nodes for storing checkpoint data from the compute nodes, the method comprising a data processor executing computer instructions of a simulation program stored in non-transitory computer readable storage medium, the simulation program including a model of the system of compute nodes and a model of the I/O subsystem, and the execution of the computer instructions performing the steps of:

(a) receiving input parameters defining the system of compute nodes and input parameters defining the I/O subsystem;

(b) computing a total number of computational floating point operations for a time between checkpoints, and computing an amount of disk storage required to store the checkpoint data for different configurations of the parallel supercomputing cluster; and

(c) presenting, to a user, a summary of the computed total number of computational floating point operations for the time between checkpoints for the different configurations of the parallel supercomputing cluster, and presenting, to the user, a summary of the computed amount of disk storage required to store the checkpoint data for the different configurations of the parallel supercomputing cluster.

10. The method as claimed in claim 9, wherein the input parameters defining the system of compute nodes include a specified checkpoint target time defining an interval of time between checkpoints.

11. The method as claimed in claim 9, wherein the different configurations of the parallel supercomputing cluster include a configuration using Lustre file system clients, running in the compute nodes.

12. The method as claimed in claim 9, wherein the different configurations of the parallel supercomputing cluster include a configuration using Parallel Log-Structured File System (PLFS) on the compute nodes on top of Lustre file system mounted on each compute node.

13. The method as claimed in claim 9, wherein the different configurations of the parallel supercomputing cluster include a configuration of the I/O subsystem using Parallel Log-Structured File System (PLFS) on the compute nodes and writing to a burst buffer that each compute node is connected to.

14. The method as claimed in claim 13, wherein the model of the I/O subsystem has checkpoint data written to the burst buffer using Message Passing Interface Input-Output (MPI-IO) during an interval of time between checkpoints when an MPI job is paused, and the burst buffer is drained to a Lustre file system during the MPI job compute time during the interval of time between checkpoints.

15. The method as claimed in claim 9, wherein step (b) includes computing an amount of time required to write checkpoint data from the compute nodes to the I/O subsystem for each of the different configurations of the parallel supercomputing cluster, and building a time series for each of the different configurations of the parallel supercomputing cluster.

16. The method as claimed in claim 9, wherein the summary of the computed total number of computational floating point operations presented to the user includes graphics of a time series of floating point operations per second and a total number of floating point operations.

17. A computer program product comprising non-transitory computer readable storage medium storing computer instructions that, when executed by a data processor, perform the steps of:

(a) receiving input parameters defining the system of compute nodes and input parameters defining the I/O subsystem;

(b) computing a total number of computational floating point operations for a time between checkpoints, and computing an amount of disk storage required to store the checkpoint data for different configurations of the parallel supercomputing cluster; and

(c) presenting, to a user, a summary of the computed total number of computational floating point operations for the time between checkpoints for the different configurations of the parallel supercomputing cluster, and presenting, to the user, a summary of the computed amount of disk storage required to store the checkpoint data for the different configurations of the parallel supercomputing cluster.

18. The computer program product as claimed in claim 17, wherein the input parameters defining the system of compute nodes include a specified checkpoint target time defining an interval of time between checkpoints, and the different configurations of the parallel supercomputing cluster include:

(1) a configuration using Lustre file system clients, running in the compute nodes;

(2) a configuration using Parallel Log-Structured File System (PLFS) on the compute nodes on top of Lustre file system mounted on each compute node; and

(3) a configuration of the I/O subsystem using PLFS on the compute nodes and writing to a burst buffer that each compute node is connected to.

19. The computer program product as claimed in claim 17, wherein the different configurations of the parallel supercomputing cluster include a configuration of the I/O subsystem using Parallel Log-Structured File System (PLFS) on the compute nodes and writing to a burst buffer that each compute node is connected to, and the model of the I/O subsystem has checkpoint data written to the burst buffer using Message Passing Interface Input-Output (MPI-IO) during an interval of time between checkpoints when an MPI job is paused, and the burst buffer is drained to a Lustre file system during the MPI job compute time during the interval of time between checkpoints.

20. The computer program product as claimed in claim 17, wherein step (b) includes computing an amount of time required to write checkpoint data from the compute nodes to the I/O subsystem for each of the different configurations of the parallel supercomputing cluster, and building a time series for each of the different configurations of the parallel supercomputing cluster.