Numerical computation algorithms for sequential checkpoint placement

Checkpoint scheduling model for optimality

“…Different from the previous [8] and [10][11][12], the proposed mathematical analytical model is independent of the explicit expression of F (t). In other words, the availability of the proposed checkpoint scheduling algorithm cannot be affected by the variety of the failure rate r(t) = F (t)…”

“…According to [9], it can be concluded that a constant CI is optimal on condition that the system fault follows Poisson/exponential process. For the particular PF 2 failure distribution, the non-increasing CI sequence can be performed in [10][11][12]. For large-scale HPC system, Liu presented the reliability-aware method for optimal checkpoint/restart strategy to minimize rollback recovery and checkpointing overheads [13,14].…”

“…For large-scale HPC system, Liu presented the reliability-aware method for optimal checkpoint/restart strategy to minimize rollback recovery and checkpointing overheads [13,14]. However, the models in [9][10][11][12][13][14] assume that no failure event occurs during the rollback recovery phase, which is not a considerate representation for the characteristic of the rollback recovery execution. Besides, [15][16][17][18][19] also intended to determine the optimal checkpoint sequence under a certain circumstance in terms of the failure distribution.…”

Comparing Checkpoint and Rollback Recovery Schemes in a Cluster System

“…Conversely, if CPs are seldom placed, a larger RB overhead after a system failure will be required. Hence, it is important to determine the optimal CP interval taking account of the trade-off between two kinds of overhead factors above [4], [5], [16]. Gelenbe at al.…”

“…Bouguerra et al [2] also give an analytical model with coordinated CP/RB for a large scale cluster system. It is worth mentioning that the above works are based on the direct application of the similar analytical techniques to the CP placement for coherent computer systems [4], [5], [16]. However, the above works did not consider the possibility of occurrence of multi-node failure.…”