A Hierarchical Approach for Dependability Analysis of a Commercial Cache-Based RAID Storage Architecture

TL;DR

This paper introduces a hierarchical simulation approach to analyze the dependability of a complex cache-based RAID storage system, focusing on fault impact, error detection, and recovery mechanisms across multiple abstraction levels.

Contribution

It develops a multi-level modeling methodology using DEPEND to evaluate fault impacts and error detection effectiveness in a commercial RAID system.

Findings

System behavior under high error rates analyzed

Error detection coverage quantified

Error accumulation in cache and disks studied

Abstract

We present a hierarchical simulation approach for the dependability analysis and evaluation of a highly available commercial cache-based RAID storage system. The archi-tecture is complex and includes several layers of overlap-ping error detection and recovery mechanisms. Three ab-straction levels have been developed to model the cache architecture, cache operations, and error detection and recovery mechanism. The impact of faults and errors oc-curring in the cache and in the disks is analyzed at each level of the hierarchy. A simulation submodel is associated with each abstraction level. The models have been devel-oped using DEPEND, a simulation-based environment for system-level dependability analysis, which provides facili-ties to inject faults into a functional behavior model, to simulate error detection and recovery mechanisms, and to evaluate quantitative measures. Several fault models are defined for each submodel to simulate cache component failures, disk failures, transmission errors, and data errors in the cache memory and in the disks. Some of the parame-ters characterizing fault injection in a given submodel cor-respond to probabilities evaluated from the simulation of the lower-level submodel. Based on the proposed method-ology, we evaluate and analyze 1) the system behavior un-der a real workload and high error rate (focusing on error bursts), 2) the coverage of the error detection mechanisms implemented in the system and the error latency distribu-tions, and 3) the accumulation of errors in the cache and in the disks.