The design for a high performance MPI implementation on the Myrinet network

Prylli, Loïc; Tourancheau, Bernard; Westrelin, Roland

doi:10.1007/3-540-48158-3_28

Cited by 31 publications

(13 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The ch self and smp plug devices are parts of the SMP implementation of MPI-BIP ( [9], [16]) and won't be further discussed in this report. Network heterogeneity is hidden by the Madeleine software layer as shown by Figure 3.…”

Section: Integrating Madeleine As An Efficient Mpich Device 41 Strucmentioning

confidence: 99%

MPICH/Madeleine: a true multi-protocol MPI for high performance networks

Aumage

Mercier

Namyst

Proceedings 15th International Parallel and Distributed Processing Symposium. IPDPS 2001

View full text Add to dashboard Cite

This report introduces a version of MPICH handling efficiently different networks simultaneously. The core of the implementation relies on a device called ch mad which is based on a generic multiprotocol communication library called Madeleine. The performance achieved with tested networks such as Fast-Ethernet, Scalable Coherent Interface or Myrinet is very good. Indeed, this multi-protocol version of MPICH generally outperforms other free or commercial implementations of MPI.Keywords: MPI,High-Performance Network,Heterogeneity,Multi-Protocol,Cluster Computing. RésuméCe rapport présente une version de MPI basée sur MPICH utilisant efficacement plusieurs réseaux simultanément. Le coeur de la mise en oeuvre repose sur un module appelé ch mad, lui-même basé sur une bibliotheèque de communication générique et multiprotocole : Madeleine. Les performances obtenues sur les réseaux testés (Fast-Ethernet, SCI, Myrinet) sont excellentes. Cette version multiprotocole de MPICH arriveà des niveaux de performanceségaux voire supérieursá ceux atteints par d'autres implémentations (libres ou commerciales) de MPI.

show abstract

Section: Integrating Madeleine As An Efficient Mpich Device 41 Strucmentioning

confidence: 99%

MPICH/Madeleine: a true multi-protocol MPI for high performance networks

Aumage

Mercier

Namyst

Proceedings 15th International Parallel and Distributed Processing Symposium. IPDPS 2001

View full text Add to dashboard Cite

show abstract

“…Iancu et al [17] examine message-strip mining for a variety of contemporary networks from the combined viewpoint of performance and portability. The idea of decomposing message traffic also appears in [24,25] where it is transparently implemented inside the network layer in order to alleviate switch contention.…”

Section: Related Workmentioning

confidence: 99%

HUNTing the overlap

Iancu

Husbands

Hargrove

2005

14th International Conference on Parallel Architectures and Compilation Techniques (PACT'05)

View full text Add to dashboard Cite

Hiding communication latency is an important optimization for parallel programs. Programmers or compilers achieve this by using non-blocking communication primitives and overlapping communication with computation or other communication operations. Using non-blocking communication raises two issues: performance and programmability. In terms of performance, optimizers need to find a good communication schedule and are sometimes constrained by lack of full application knowledge. In terms of programmability, efficiently managing nonblocking communication can prove cumbersome for complex applications. In this paper we present the design principles of HUNT, a runtime system designed to search and exploit some of the available overlap present at execution time in UPC programs. Using virtual memory support, our runtime implements demand-driven synchronization for data involved in communication operations. It also employs message decomposition and scheduling heuristics to transparently improve the non-blocking behavior of applications. We provide a user level implementation of HUNT on a variety of modern high performance computing systems. Results indicate that our approach is successful in finding some of the overlap available at execution time. While system and application characteristics influence performance, perhaps the determining factor is the time taken by the CPU to execute a signal handler. Demand driven synchronization at execution time eliminates the need for the explicit management of non-blocking communication. Besides increasing programmer productivity, this feature also simplifies compiler analysis for communication optimizations.

show abstract

“…Every OS-bypass MPI implementation described in current literature [15,29,25,11] requires application processing to move data. These implementations typically use a two-level protocol, where short messages are sent eagerly and long messages are sent using a rendezvous protocol.…”

Section: Mpi Progressmentioning

confidence: 99%