Distributed applications running on a large cluster environment, such as the cloud instances will have shorter execution time. However, the application might suffer from sudden termination due to unpredicted computing node failures, thus loosing the whole computation. Checkpoint/restart is a fault tolerance technique used to solve this problem. In this work we evaluated the performance of two of the most commonly used checkpoint/restart techniques (Distributed Multithreaded Checkpointing (DMTCP) and Berkeley Lab Checkpoint/Restart library (BLCR) integrated into the OpenMPI framework). We aimed to test their validity and evaluate their performance in both local and Amazon Elastic Compute Cloud (EC2) environments. The experiments were conducted on Amazon EC2 as a well-known proprietary cloud computing service provider. Results obtained were reported and compared to evaluate checkpoint and restart time values, data scalability and compute processes scalability. The findings proved that DMTCP performs better than BLCR for checkpoint and restart speed, data scalability and compute processes scalability experiments.

翻译：分布式应用在大型集群环境（如云实例）中运行时，执行时间通常较短。然而，应用可能因不可预测的计算节点故障而突然终止，导致整个计算损失。检查点/重启是一种用于解决该问题的容错技术。本研究评估了两种最常用的检查点/重启技术（分布式多线程检查点（DMTCP）和集成到OpenMPI框架中的伯克利实验室检查点/重启库（BLCR））的性能。我们旨在测试其有效性，并在本地环境和亚马逊弹性计算云（EC2）环境中评估其性能。实验在知名专有云计算服务提供商亚马逊EC2上进行。我们报告并比较了实验结果，以评估检查点与重启的时间值、数据可扩展性及计算进程可扩展性。研究结果表明，在检查点与重启速度、数据可扩展性及计算进程可扩展性实验中，DMTCP的性能优于BLCR。