Parallel Scaling Characteristics of Selected NERSC User ProjectCodes

Skinner, David; Verdier, Francesca; Anand, Harsh; Carter, Jonathan; Durst, Mark; Gerber, R.

doi:10.2172/885226

Search citation statements

Order By: Relevance

Paper Sections

Select...

System-ranked Process Groups1

Citation Types

Supporting

Mentioning

Contrasting

Year Published

2012

Publication Types

Select...

Book1

Relationship

Self Cite0

Independent1

Authors

Journals

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The results of these collectives are independent of the ranks from which the individual data contributions arise (assuming commutative operations). There is also evidence that a sizable fraction of collective communication in applications involve these operations [1,13,15]. We enumerate a few examples of algorithms and applications that use just these collectives:…”

Section: System-ranked Process Groupsmentioning

confidence: 99%

Scalable Algorithms for Constructing Balanced Spanning Trees on System-Ranked Process Groups

Langer

Venkataraman

Kalé

2012

Recent Advances in the Message Passing Interface

View full text Add to dashboard Cite

Abstract. Current implementations of process groups (subcommunicators) have non-scalable (O(group size)) memory footprints and even worse time complexities for setting up communication. We propose systemranked process groups, where member ranks are picked by the runtime system, as a cheaper and faster alternative for a subset of collective operations (barrier, broadcast, reduction, allreduce). This paper presents two distributed algorithms for balanced, k-ary spanning tree construction over system-ranked process groups obtained by splitting a parent group. Our schemes have much smaller memory footprints and also perform better, even at modest process counts. We demonstrate performance results up to 131, 072 cores of BlueGene/P.

show abstract

Section: System-ranked Process Groupsmentioning

confidence: 99%