PROTEUS: A PORTABLE SIMULATION PROGRAM FOR ESTIMATING DATA LOSS RISKS IN DISK ARRAYS

Abstract

Providing trustworthy estimates of the reliability of fault-tolerant disk arrays is a hard task because analytical techniques are based on assumptions that are never realized in actual disk arrays and simulation techniques require writing a new simulation program for each array organization we want to investigate. We wrote the Proteus simulation program to address these issues. First, Proteus is flexible and can be parameterized to model most fault-tolerant disk array organizations. Second, Proteus designed to run fast, which is important because obtaining tight confidence intervals for the reliability of highly fault-tolerant disk arrays often requires millions of simulation runs. Finally, Proteus is written in Python 3, a freely available language that has been ported to many programming environments. We used Proteus to evaluate the reliability of various fault-tolerant disk array organizations, among which RAID levels 4, 5, and 6 and two-dimensional RAID arrays. Our measurements have shown a surprisingly good agreement with the reliability figures obtained through analytical techniques and indicate that there is very little difference between the reliability values obtained assuming deterministic repair times and those assuming exponential repair times.

Description

Keywords

Proteus, RAID, Simulations, Python, Portable, Estimating, Data loss, Disk array

Citation