Using Replication for Resilience on Exascale Systems

Abstract

High performance computing applications must be tolerant to faults, which are common occurrences especially in post-petascale settings. The traditional fault-tolerance solution is checkpoint-rollback, by which the application saves its state to secondary storage throughout execution and recover from the latest saved state in case of a failure. An oft studied research question is that of the optimal checkpointing strategy: when should checkpoints be saved. Unfortunately, even using an optimal checkpointing strategy, the frequency of checkpointing must increase as platform scale increases, leading to higher checkpointing overhead. This overhead precludes high parallel efficiency for large-scale platforms, thus mandating other more scalable fault-tolerance mechanisms. One such mechanism is replication, which can be used in addition to checkpoint-rollback. Using replication, multiple processors perform the same computation so that a processor failure does not necessarily mean application failure. While at first glance replication may seem wasteful, it may be significantly more efficient than using solely checkpointrollback at large scale. In this work we investigate two approaches for replication. In the first approach, each process in a single instance of a parallel application is (transparently) replicated. In the second approach, entire application instances are replicated. We provide a theoretical study of these two approaches, comparing them to checkpoint-rollback only, in terms of expected application execution time.

Adaptive and Power-Aware Resilience for Extreme-scale Computing

“…We use the term core to represent the computing resource allocation unit [28]. We further use P (σ, w, t) to denote a process executing at rate σ to complete a workload w by time t. The basic tenet of Lazy Shadowing is the concept of shadowing, whereby each process is associated with a lazy replica.…”

“…The impact of process replication on MNFTI has been studied in [28]. Our problem is equivalent to that, with the difference that our work can tolerate one failure in each shadowed set while [28] can tolerate one failure in each replica-group of size 2.…”

“…Our problem is equivalent to that, with the difference that our work can tolerate one failure in each shadowed set while [28] can tolerate one failure in each replica-group of size 2. Therefore, we can apply the methodology in [28] to our case, and the MNFTI with Lazy Shadowing for different number of shadowed sets (S) is shown in Table 1. Note that when processes are not replicated, every failure would interrupt the application, i.e., MNFTI=1, so MNFTI can be significantly increased by Lazy Shadowing.…”

Resilient N-Body Tree Computations with Algorithm-Based Focused Recovery: Model and Performance Analysis

“…Replication remains the most transparent and least intrusive technique and can be used at different levels (duplication, triplication or even more) . Combined with checkpointing, replication comes with two flavors: process replication [24,25] and group replication [26]. Process replication applies to message-passing applications with communicating processes.…”

Coping with silent and fail-stop errors at scale by combining replication and checkpointing

“…When transparent replication is not (yet) provided by the runtime system, one solution could be to implement it explicitly within the application, but this is a labor-intensive process especially for legacy applications. Another approach introduced in [15] is group replication, a technique that can be used whenever process replication is not available. Group replication is agnostic to the parallel programming model, and thus views the application as an unmodified black box.…”