Zero-copy protocol for MPI using infiniband unreliable datagram

Koop, Matthew J.; Sur, Sayantan; Panda, Dhabaleswar K.

doi:10.1109/clustr.2007.4629230

Cited by 13 publications

(1 citation statement)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…UD is a connection-less and unreliable transportmessages may be lost during transmission but a single endpoint can communicate with any other endpoint in the system. UD-based MPI implementations can reduce memory usage up to 30 times compared to RC-based implementations [2,5]. Even though a UD-based MPI implementation provides significant scalability advantages over an RC-based implementation, there is an added cost at the MPI layer due to message retransmissions when network errors occur (MPI is a reliable transport).…”

Section: Introductionmentioning

confidence: 99%

A study of application-level recovery methods for transient network faults

Laguna

León

Schulz

et al. 2013

Proceedings of the Workshop on Latest Advances in Scalable Algorithms for Large-Scale Systems

View full text Add to dashboard Cite

With the increasing number of components in HPC systems, transient faults will become commonplace. Today, network transient faults, such as lost or corrupted network packets, are addressed by middleware libraries at the cost of high memory usage and packet retransmissions. These costs, however, can be eliminated using application-level fault tolerance. In this paper, we propose recovery methods for transient network faults at the application level. These methods reconstruct missing or corrupted data via interpolation. We derive a realistic fault model using network fault rates from a production HPC cluster and use it to demonstrate the effectiveness of our reconstruction methods in an FFT kernel. We found that the normalized root-mean-square error for FFT computations can be as low as 0.1% and, thus, demonstrates that network faults can be handled at the application level with low perturbation in applications that have smoothness in their computed data.

show abstract

Section: Introductionmentioning

confidence: 99%