Self-stabilization Overhead: an Experimental Case Study on Coded Atomic Storage

Abstract: Shared memory emulation can be used as a fault-tolerant and highly available distributed storage solution or as a low-level synchronization primitive. Attiya, Bar-Noy, and Dolev were the first to propose a single-writer, multi-reader linearizable register emulation where the register is replicated to all servers. Recently, Cadambe et al. proposed the Coded Atomic Storage (CAS) algorithm, which uses erasure coding for achieving data redundancy with much lower communication cost than previous algorithmic solutions. Although CAS can tolerate server crashes, it was not designed to recover from unexpected, transient faults, without the need of external (human) intervention. In this respect, Dolev, Petig, and Schiller have recently developed a self-stabilizing version of CAS, which we call CASSS. As one would expect, self-stabilization does not come as a free lunch; it introduces, mainly, communication overhead for detecting inconsistencies and stale information. So, one would wonder whether the overhead introduced by self-stabilization would nullify the gain of erasure coding. To answer this question, we have implemented and experimentally evaluated the CASSS algorithm on PlanetLab; a planetary scale distributed infrastructure. The evaluation shows that our implementation of CASSS scales very well in terms of the number of servers, the number of concurrent clients, as well as the size of the replicated object. More importantly, it shows (a) to have only a constant overhead compared to the traditional CAS algorithm (which we also implement) and (b) the recovery period (after the last occurrence of a transient fault) is as fast as a few client (read/write) operations. Our results suggest that CASSS does not significantly impact efficiency while dealing with automatic recovery from transient faults and bounded size of needed resources.