Enhancing Checkpoint Performance with Staging IO and SSD

Ouyang, Xiangyong; Marcarelli, Sonya; Panda, Dhabaleswar K.

doi:10.1109/snapi.2010.10

Cited by 27 publications

(14 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Their approach can be incorporated with the semi-blocking algorithm by relaxing the stall of computation when taking global checkpoint. Ouyang et al [16] enhance the performance of checkpointing to SSD with aggregation of checkpoint buffer and staging IO. Our approach is different from theirs in that checkpoints are distributed among the SSD disks rather than stored in a central checkpoint server.…”

Section: Related Workmentioning

confidence: 99%

Hiding Checkpoint Overhead in HPC Applications with a Semi-Blocking Algorithm

Meneses

Kalé

2012

2012 IEEE International Conference on Cluster Computing

View full text Add to dashboard Cite

Abstract-The HPC community has seen a steady increase in the number of components in every generation of supercomputers. Assembling a large number of components into a single cluster makes a machine more powerful, but also much more prone to failures. Therefore, fault tolerance has become a major concern in HPC. To deal with node crashes in large systems, checkpoint/restart is by far the preferred method. A typical way to implement checkpoints is by using a blocking algorithm, which suspends the execution of the application while the checkpoint is safely stored. One limitation of the blocking algorithm is that it saturates the network bandwidth at the time of checkpoint. This problem will become even more critical because the projected network bandwidth increase will not match the increase in memory per node. To alleviate this problem, we have developed a semi-blocking checkpoint algorithm that overlaps execution of the application with transmission of checkpoints. Our implementation decomposes a checkpoint into small messages that are interleaved with application messages. The experimental results show a dramatic reduction in the checkpoint overhead for various applications. We present a model for our approach and use this model to compute the benefit of the semi-blocking algorithm for different failure rates predicted at Exascale. We estimate our method can reduce up to 22% the total execution time of an iterative scientific application.

show abstract

Section: Related Workmentioning

confidence: 99%

Hiding Checkpoint Overhead in HPC Applications with a Semi-Blocking Algorithm

Meneses

Kalé

2012

2012 IEEE International Conference on Cluster Computing

View full text Add to dashboard Cite

show abstract

“…Further enhancements with dynamic buffering have been carried out in [70] that use a small amount of buffer pool to accelerate checkpoint operations further. These schemes have been evaluated in conjunction with a new staging I/O framework and modern Solid State Disks (SSD) to provide a high-performance Checkpoint-Restart solution [72].…”

Section: End-to-end Reliable Data Transmission In Mvapichmentioning

confidence: 99%

“…Of particular note are (1) the work done by the MVAPICH team at Ohio State University to accelerate checkpoint I/O using buffering and (2) the hook interface used by Open MPI's parallel debugger support. In the first case, the work done by the MVAPICH team [70][71][72]96] influenced a rewrite of the I/O code within BLCR that will appear in the next BLCR release and deliver the level of I/O performance "out of the box" that currently requires use of a buffering agent such as that implemented by the MVAPICH team. In the second case, the work done by the Open MPI team at Indiana University (IU) on integrating MPI and checkpoint/restart with a parallel debugger identified a need to allow the debugger support to interact with the checkpointing and restarting mechanism in ways not previously envisioned by the BLCR team.…”

Section: Blcr-based Checkpoint/restart Support In Mpi Implementationsmentioning

confidence: 99%

Coordinated Fault Tolerance for High-Performance Computing

Dongarra¹,

Bosilca²

2013

View full text Add to dashboard Cite

“…FlashCache [9] uses Àash memory as a secondary ¿le cache for web servers. Lee et al [10] analyze the impact of using SSDs in transaction processing, while [11] examines how SSDs can improve checkpointing performance. In [12] is examined how SSDs can be used as a large cache on top of RAID to conserve energy.…”

Section: Related Workmentioning

confidence: 99%

Analysis of Prediction and Replacement Algorithms Applied to Real Workload for Storage Devices

Sette

Cartaxo

Chiu

et al. 2012

2012 IEEE 20th International Symposium on Modeling, Analysis and Simulation of Computer and Telecommunication Systems

View full text Add to dashboard Cite

This work evaluates cache algorithms for block devices in terms of hit rate for different replacement and prediction algorithms, and also focuses on choosing suitable replacement and prediction strategies to be implemented in a storage array caching solution employing solid-state drives. As case study, a real workload was continuously collected from web proxy server network. Such replacement and prediction algorithms were evaluated in detail and compared between them considering traced SCSI commands collected from Linux kernel. Comparison results for replacement algorithms LRU, CLOCK, LRFU and LRU-WAR applied to real workload indicates that LRU obtained good results in the majority of the cases. GHB and ReadAhead prefetching algorithms also were integrated and improvements of about 5.5% were achieved when ReadAhead was used.

show abstract

Enhancing Checkpoint Performance with Staging IO and SSD

Cited by 27 publications

References 13 publications

Hiding Checkpoint Overhead in HPC Applications with a Semi-Blocking Algorithm

Hiding Checkpoint Overhead in HPC Applications with a Semi-Blocking Algorithm

Coordinated Fault Tolerance for High-Performance Computing

Analysis of Prediction and Replacement Algorithms Applied to Real Workload for Storage Devices

Contact Info

Product

Resources

About