Asynchronous MPI messaging on Myrinet

Keppitiyagama, Chamath; Wagner, Alan

doi:10.1109/ipdps.2001.924989

Cited by 11 publications

(12 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Various techniques have been developed to overcome these problems. The techniques can be broadly classified into three types: using interrupts to improve communication progress [1], [17], [19], using asynchronous communication progress to improve communication-computation overlaps [8], [9], [10], [12], [20], and improving the protocol design [3], [15], [16], [18], [19]. The interrupt driven message detection approach [1], [17], [19] allows each party (sender or receiver) to react to a message whenever the message arrives.…”

Section: Related Workmentioning

confidence: 99%

“…Asynchronous communication progress allows communications to be performed asynchronously with the main computation thread. This approach either needs a helper thread [10], [12], [20] or requires additional hardware support [8], [9]. Allowing communication and computation overlaps with a helper thread incurs performance penalties for synchronous communications.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Near-Optimal Rendezvous Protocols for RDMA-Enabled Clusters

Small

Yuan

2010

2010 39th International Conference on Parallel Processing

View full text Add to dashboard Cite

Abstract-Optimizing Message Passing Interface (MPI) pointto-point communication for large messages is of paramount importance since most communications in MPI applications are performed by such operations. Remote Direct Memory Access (RDMA) allows one-sided data transfer and provides great flexibility in the design of efficient communication protocols for large messages. However, achieving high performance on RDMAenabled clusters is still challenging due to the complexity both in communication protocols and in protocol invocation scenarios. In this work, we investigate a profile-driven compiled-assisted protocol customization approach for efficient communication on RDMA-enabled clusters. We analyze existing protocols and show that they are not ideal in many situations. By leveraging the RDMA capability, we develop a set of protocols that can provide near-optimal performance for all protocol invocation scenarios, which allows protocol customization to achieve near-optimal performance when the appropriate protocol is used for each communication. Finally, we evaluate the potential benefits of protocol customization using micro-benchmarks and application benchmarks. The results demonstrate that the proposed protocols can out-perform traditional rendezvous protocols to a large degree in many situations and that protocol customization can significantly improve MPI communication performance.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Near-Optimal Rendezvous Protocols for RDMA-Enabled Clusters

Small

Yuan

2010

2010 39th International Conference on Parallel Processing

View full text Add to dashboard Cite

show abstract

“…[1,5]) and various schemes that improve the communication progress and the communicationcomputation overlap capability have been proposed. Existing techniques can in general be classified into three types: using interrupts to improve communication progress [1,17,18], using asynchronous communication progress to improve communication-computation overlaps [6,7,8,10,19], and improving the protocol design [2,15,16,18]. The interrupt driven message detection approach [1,17,18] allows each party (sender or receiver) to react to a message whenever the message arrives.…”

Section: Related Workmentioning

confidence: 99%

“…The asynchronous communication progress allows communications to be performed asynchronously with the main computation thread. This approach either needs a helper thread [8,10,19] or requires additional hardware support [6,7]. This approach focuses on improving communication and computation overlaps and often introduces penalties for synchronous communications.…”

Section: Related Workmentioning

confidence: 99%

“…We will discuss the subtleties next. Poll control messages and eager messages (3) If (receiving a matching eager message) (4) Copy eager message to the user buffer; return; (5) If (receiving a matching SENDER READY message) (6) If (message size is less than HYBRID THRESHOLD) (7) Issue RDMA read to get the data; (8) wait for the read to complete; return (9) Else drop the message /* useless message */ (10) If (receiving a matching FIN message) return; (11) Goto (2); /* loop until one of the above happens */…”

Section: Integrating the Four Protocolsmentioning

confidence: 99%

See 1 more Smart Citation

Maximizing MPI point-to-point communication performance on RDMA-enabled clusters with customized protocols

Small

Yuan

2009

Proceedings of the 23rd International Conference on Supercomputing

View full text Add to dashboard Cite

Message Passing Interface (MPI) point-to-point communications are usually realized with two protocols, the eager protocol for small messages and the rendezvous protocol for medium and large sized messages. Traditional senderinitiated rendezvous protocols are sub-optimal in many situations. In this work, we propose to refine the rendezvous protocol for medium and large messages on RDMA-enabled clusters with three protocols that are customized for different situations, a hybrid protocol for medium sized messages when the sender arrives early, a sender-initiated protocol for large messages when the sender arrives early, and a receiver-initiated protocol when the receiver arrives early. In comparison to traditional sender-initiated rendezvous protocols, the proposed scheme reduces unnecessary synchronizations, decreases the number of control messages that are in the critical path of communications, and improves the communication progress, which results in a significantly better communication-computation overlap capability. We present and analyze these protocols, and describe how these protocols and the eager protocol can be seamlessly integrated in one system without introducing an excessive number of control messages. We have implemented the proposed scheme for InfiniBand clusters. The experimental results demonstrate the effectiveness of the proposed technique.

show abstract

Support for MPI at the Network Interface Level

Tourancheau¹,

Westrelin

2001

Recent Advances in Parallel Virtual Machine and Message Passing Interface

View full text Add to dashboard Cite

Asynchronous MPI messaging on Myrinet

Cited by 11 publications

References 9 publications

Near-Optimal Rendezvous Protocols for RDMA-Enabled Clusters

Near-Optimal Rendezvous Protocols for RDMA-Enabled Clusters

Maximizing MPI point-to-point communication performance on RDMA-enabled clusters with customized protocols

Support for MPI at the Network Interface Level

Contact Info

Product

Resources

About