Improving Scalability of Application-Level Checkpoint-Recovery by Reducing Checkpoint Sizes

Abstract

The execution times of large-scale parallel applications on nowadays multi/many-core systems are usually longer than the mean time between failures. Therefore, parallel applications must tolerate hardware failures to ensure that not all computation done is lost on machine failures. Checkpointing and rollback recovery is one of the most popular techniques to implement fault-tolerant applications. However, checkpointing parallel applications is expensive in terms of computing time, network utilization and storage resources. Thus, current checkpoint-recovery techniques should minimize these costs in order to be useful for large scale systems. In this paper three different and complementary techniques to reduce the size of the checkpoints generated by application-level checkpointing are proposed and implemented. Detailed experimental results obtained on a multicore cluster show the effectiveness of the proposed methods to reduce checkpointing cost.

Time-stamp incremental checkpointing and its applying for an optimization of execution model to improve performance of CAPE

“…However, the checkpoint's size is too large. Incremental checkpointing [11,21,22,23,12,24] only saves the modified information as compared to the previous checkpoint. This technique reaches advantages of reducing checkpoint's overhead and checkpoint's size, so it is in widely used in distributed computing.…”

“…This technique reaches advantages of reducing checkpoint's overhead and checkpoint's size, so it is in widely used in distributed computing. Besides, using data compression to reduce checkpoint's size [11,21,24], it is also focus on the techniques that detect modified data but reach the minimum of size. Some typical techniques are using page-based protection to identify the pages in memory that have been modified [11,22,23], using word-level granularity [21,12], using block encoding [22], using user-directed and memory exclusion [11], using live variable analysis [24].…”

“…Besides, using data compression to reduce checkpoint's size [11,21,24], it is also focus on the techniques that detect modified data but reach the minimum of size. Some typical techniques are using page-based protection to identify the pages in memory that have been modified [11,22,23], using word-level granularity [21,12], using block encoding [22], using user-directed and memory exclusion [11], using live variable analysis [24]. In CAPE, Discontinuous Incremental Checkpointing (DICKPT) is a development based on incremental checkpointing, that contains two kinds of data, register information and modified data of the process.…”

Two-Level Incremental Checkpoint Recovery Scheme for Reducing System Total Overheads

“…To further reduce the overheads of the checkpoint recovery process, Cores et al . [28] carry out the study on how to reduce the size of the checkpoint files. Also, for large-scale distributed systems, Wei et al .…”

Resilience and fault tolerance in high-performance computing for numerical weather and climate prediction

“…We refer to the literature (e.g. Bautista-Gomez and Cappello, 2015;Berrocal et al, 2015Berrocal et al, , 2016Turnbull and Alldrin, 2003) for fault detection and prediction, as well as for other techniques, such as advanced focused checkpointing strategies (Cores et al, 2013;Islam et al, 2013;Kohl et al, 2019;Losada et al, 2019;Plank et al, 1995;Rodríguez et al, 2010;Sancho et al, 2004;Tao et al, 2018), selective identification of parts of models in need of reliability (Bridges et al, 2012;Hoemmen et al, 2011), self-stabilizing iterative solvers (Sao and Vuduc, 2013), data compression techniques (Di and Cappello, 2016), global view resilience (Chien et al, 2015), and resilience in the framework of domain decomposition preconditioners (Rizzi et al, 2018a(Rizzi et al, , 2018bSargsyan et al, 2015).…”