Title :
High Performance Information Management for HPC Parallel Computing
Author :
Tucker, Scot ; Spetka, Scott ; Ramseyer, George ; Emeny, Susan ; Fitzgerald, Dennis ; Linderman, Richard
Author_Institution :
ITT Corp., Rome, NY
Abstract :
A new pub/sub/query (PSQ) paradigm has been developed for the information management of high performance computers. The main advantage of PSQ over Message Passing Interface (MPI) is the ability to adapt to dynamically changing requirements. Information management for high performance computing (HPC) applications also coordinates the distribution of data across processing nodes, both for initial data and for intermediate results distribution. Additionally, PSQ exhibits enhanced robustness. A matrix multiplication case study illustrates the advantages of the PSQ software development approach.
Keywords :
information management; parallel programming; software engineering; HPC parallel computing; high performance computers; high performance information management; message passing interface; pub-sub-query paradigm; software development; Checkpointing; Computer crashes; High performance computing; Information management; Laboratories; Master-slave; Parallel processing; Programming environments; Robustness; Subscriptions;
Conference_Titel :
DoD HPCMP Users Group Conference, 2008. DOD HPCMP UGC
Conference_Location :
Seattle, WA
Print_ISBN :
978-1-4244-3323-0
DOI :
10.1109/DoD.HPCMP.UGC.2008.67