NFuSA - Neuro-Fuzzy Algorithm for Sparing in RAID Systems

Sparing, the process of rebuilding data in case of disk failure, has been a target of research since early 1990´s. The problem that these specific hardware/software control systems typically face in sparing is the tradeoff between serving requests - user´s versus internal. If the algorithm favors user requests, in the presence of heavy workloads, the internal data recovery gets preempted resulting in risky delay of the data sparing. On the other hand, favoring internal data recovery requests over the user requests can result in high response times per transaction that are unacceptable for the users of the RAID system. Intelligent, neuro-fuzzy controllers (NFCs) offer a way to improve the control process and enhance the ability of a system to achieve faster system response, while serving the internal requests at the same time. This paper presents the neuro-fuzzy enhancement of the traditional data recovery of a RAID system modeled with a queue system with vacations (QSV). Experimental results demonstrated better balancing between an acceptable response time for the user requests and the time for the data to be redundant again, resulting in both higher user satisfaction and better system reliability.