Optimizing a conjugate gradient solver with non-blocking collective operations

Hoefler, Torsten; Gottschling, Peter; Lumsdaine, Andrew; Rehm, Wolfgang

doi:10.1016/j.parco.2007.06.006

Cited by 54 publications

(23 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We therefore conclude that overlapping the neighbor exchange communication with steps 2 and 3 should show a reasonable performance benefit at any scale. Overlapping this kind of communication has been successfully demonstrated on a regular grid in [1]. We expect the irregular grid to achieve similar performance improvements which could result in a reduction of the communication overhead.…”

Section: Domain Parallelizationmentioning

confidence: 87%

See 1 more Smart Citation

Sparse Non-blocking Collectives in Quantum Mechanical Calculations

Hoefler

Lorenzen

Lumsdaine

2008

Recent Advances in Parallel Virtual Machine and Message Passing Interface

Self Cite

View full text Add to dashboard Cite

Abstract. For generality, MPI collective operations support arbitrary dense communication patterns. However, in many applications where collective operations would be beneficial, only sparse communication patterns are required. This paper presents one such application: Octopus, a production-quality quantum mechanical simulation. We introduce new sparse collective operations defined on graph communicators and compare their performance to MPI Alltoallv. Besides the scalability improvements to the collective operations due to sparsity, communication overhead in the application was reduced by overlapping communication and computation. We also discuss the significant improvement to programmability offered by sparse collectives.

show abstract

Section: Domain Parallelizationmentioning

confidence: 87%

“…The basic equations of DFT are the static and time-dependent KohnSham equations: 1 Hϕ j = ε j ϕ j i ∂ ∂t ϕ j (t) = H(t)ϕ j (t)…”

Section: Introductionmentioning

confidence: 99%

Sparse Non-blocking Collectives in Quantum Mechanical Calculations

Hoefler

Lorenzen

Lumsdaine

2008

Recent Advances in Parallel Virtual Machine and Message Passing Interface

Self Cite

View full text Add to dashboard Cite

show abstract

“…Examples include [29] and can be found at the LibNBC webpage [30]. Other double-buffering based schemes to optimize parallel implementations of more algorithms (e.g.…”

Section: Discussionmentioning

confidence: 99%

A Case for Non-blocking Collective Operations

Hoefler

Squyres

Rehm

et al. 2006

Frontiers of High Performance Computing and Networking – ISPA 2006 Workshops

Self Cite

View full text Add to dashboard Cite

Abstract. Non-blocking collective operations for MPI have been in discussion for a long time. We want to contribute to this discussion and to give a rationale for the usage these operations and assess their possible benefits. A LogGP model for the CPU overhead of collective algorithms and a benchmark to measures it are provided and show a large potential to overlap communication and computation. We show that nonblocking collective operations can provide at least the same benefits as non-blocking point to point operations already do. Our claim is that actual CPU overhead for non-blocking collective operations depends on the message size and the communicator size and benefits especially highly scalable applications with huge communicators. We prove that the share of the overhead of the overall communication time of current blocking collective operations gets smaller with bigger communicators and larger messages. We show that the user level CPU overhead is less than 10% for MPICH2 and LAM/MPI using TCP/IP communication, which leads us to the conclusion that, by using non-blocking collective communication, ideally 90% idle CPU time can be freed for the application.

show abstract

“…A particular example with a three-dimensional Poisson equation showed a performance improvement of 34% by applying nonblocking collective operations [5]. The runtime of a strong-scaling medical image reconstruction algorithm [6] could be improved up to 8%.…”

Section: Nonblocking Collective Communicationmentioning

confidence: 99%