Experiences Parallelizing, Configuring, Monitoring and Visualizing Applications for Clusters and Multi-Clusters

Publisher Summary This chapter provides an overview of the PATHS system. The PATHS system uses a “wrapper” to provide a level of indirection to the actual run-time location of data making the data available from wherever threads or processes are located. A path is comprised of one or more wrappers. A performance gain of 1.52, 1.79, and 1.98 is achieved by reconfiguring the LAM-MPI Allreduce operation. The performance of the unmodified Allreduce operation when using two clusters interconnected by a WAN link with 30-50ms roundtrip latency is measured. Configurations that resulted in multiple messages being sent across the WAN did not add any significant performance penalty to the unmodified Allreduce operation for packet sizes up to 4KB. For larger packet sizes, the Allreduce operation rapidly deteriorated performance wise. To log and visualize the performance data the chapter discusses event space, a configurable data collecting, management, and observation system used for monitoring low-level synchronization and communication behavior of parallel applications on clusters and multi-clusters. The chapter discusses examples on how the data in a virtual event space can be used for analyzing the communication behavior of the wind tunnel parallel application.

[1]  Richard Wolski,et al.  The network weather service: a distributed resource performance forecasting service for metacomputing , 1999, Future Gener. Comput. Syst..

[2]  Andrea C. Arpaci-Dusseau,et al.  Searching for the sorting record: experiences in tuning NOW-Sort , 1998, SPDT '98.

[3]  Calton Pu,et al.  Infopipes: an Abstraction for Multimedia Streaming , 2002 .

[4]  G. Amdhal,et al.  Validity of the single processor approach to achieving large scale computing capabilities , 1967, AFIPS '67 (Spring).

[5]  William E. Johnston,et al.  The NetLogger methodology for high performance distributed systems performance analysis , 1998, Proceedings. The Seventh International Symposium on High Performance Distributed Computing (Cat. No.98TB100244).

[6]  Dean Sutherland,et al.  The architecture of the Remos system , 2001, Proceedings 10th IEEE International Symposium on High Performance Distributed Computing.

[7]  Brian Vinter,et al.  Scalable processing and communication performance in a multi-media related context , 2002, Proceedings. 28th Euromicro Conference.

[8]  Daniel A. Reed,et al.  Virtual Reality and Parallel Systems Performance Analysis , 1995, Computer.

[9]  Otto J. Anshus,et al.  Configurable Collective Communication in LAM-MPI , 2002 .

[10]  Brian Vinter,et al.  PATHS - Integrating the Principles of Method-Combination and Remote Procedure Calls for Run-Time Configuration and Tuning of High-Performance Distributed Applications YYYY No org found YYY , 2001 .

[11]  Jeffrey S. Vetter,et al.  Autopilot: adaptive control of distributed applications , 1998, Proceedings. The Seventh International Symposium on High Performance Distributed Computing (Cat. No.98TB100244).

[12]  Liviu Iftode,et al.  Monitoring shared virtual memory performance on a Myrinet-based PC cluster , 1998, ICS '98.

[13]  Brian Vinter,et al.  Java PastSet: a structured distributed shared memory system , 2003, IEE Proc. Softw..

[14]  Steve Sistare,et al.  MPI support in the Prism programming environment , 1999, SC '99.

[15]  Greg Burns,et al.  LAM: An Open Cluster Environment for MPI , 2002 .

[16]  Jack J. Dongarra,et al.  Review of Performance Analysis Tools for MPI Parallel Programs , 2001, PVM/MPI.

[17]  Message P Forum,et al.  MPI: A Message-Passing Interface Standard , 1994 .

[18]  Jonathan Walpole,et al.  Gscope: A Visualization Tool for Time-Sensitive Software , 2002, USENIX Annual Technical Conference, FREENIX Track.

[19]  J. L. Traff.  Implementing the MPI Process Topology Mechanism , 2002, ACM/IEEE SC 2002 Conference (SC'02).

[20]  John Markus Bjørndalen,et al.  EventSpace - Exposing and Observing Communication Behavior of Parallel Cluster Applications , 2003, Euro-Par.

[21]  John L. Gustafson,et al.  Reevaluating Amdahl's law , 1988, CACM.

[22]  Evgenia Smirni,et al.  I/O, performance analysis, and performance data immersion , 1996, Proceedings of MASCOTS '96 - 4th International Workshop on Modeling, Analysis and Simulation of Computer and Telecommunication Systems.

[23]  Ruth A. Aydt,et al.  A Grid Monitoring Architecture , 2002 .

[24]  William L. George,et al.  The Interoperable Message Passing Interface (IMPI) Extensions to LAM/MPI , 2001 .

[25]  Brian Vinter,et al.  EXTENDING THE APPLICABILITY OF SOFTWARE DSM BY ADDING USER REDEFINABLE MEMORY SEMANTICS , 2002 .

[26]  Jeffrey S. Vetter,et al.  An Empirical Performance Evaluation of Scalable Scientific Applications , 2002, ACM/IEEE SC 2002 Conference (SC'02).

[27]  Message Passing Interface Forum.  MPI: A message - passing interface standard , 1994 .

[28]  Jason Lee,et al.  A Monitoring Sensor Management System for Grid Environments , 2000, Proceedings the Ninth International Symposium on High-Performance Distributed Computing.

[29]  Dhabaleswar K. Panda.  Issues in Designing Efficient and Practical Algorithms for Collective Communication on Wormhole-Rout , 1995 .

[30]  John Markus Bjørndalen,et al.  Using a virtual event space to understand parallel application communication behavior , 2003 .