I/O Forwarding for Quiet Clusters

Jaka Močnik (), Marko Novak () and Erich Focht ()
Additional contact information
Jaka Močnik: XLAB d.o.o.
Marko Novak: XLAB d.o.o.
Erich Focht: NEC HPC Europe

A chapter in High Performance Computing on Vector Systems 2011, 2011, pp 21-39 from Springer

Abstract: Abstract One of the important factors limiting the scalability of parallel applications to large-scale computing systems is the noise of the operating system running on the compute nodes. In this paper, the concept of OS noise is presented, noise sources are listed, and impact on (parallel) applications discussed. State-of-the-art means for noise mitigation are described. Two well-established synthetic micro-benchmarks for measuring noise are presented and their results on a test system are given. Noise induced by I/O operations is studied in detail and measured for different I/O software technologies. Design and implementation of an application-level framework for forwarding I/O operations to a dedicated I/O server, designed specifically to reduce the I/O-induced noise, is presented and compared to existing I/O technologies. The results show that the approach reduces noise significantly when compared even to state-of-the-art I/O solutions while maintaining good performance, but at a cost of sacrificing advanced file system functionalities. The developed software components used for implementing I/O forwarding enable users to achieve higher scalability on commodity based clusters with Infiniband interconnect with methods that were until now available only on a few specially designed HPC parallel computers.

Keywords: IEEE Computer Society; Noise Source; High Performance Computing; Parallel Application; Kernel Thread (search for similar items in EconPapers)
Date: 2011
References: Add references at CitEc
Citations:

There are no downloads for this item, see the EconPapers FAQ for hints about obtaining it.

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:spr:sprchp:978-3-642-22244-3_2

Ordering information: This item can be ordered from
http://www.springer.com/9783642222443

DOI: 10.1007/978-3-642-22244-3_2

Access Statistics for this chapter

More chapters in Springer Books from Springer
Bibliographic data for series maintained by Sonal Shukla () and Springer Nature Abstracting and Indexing ().