The Clustered Raid 5 (CRAID5) architecture with a parity group size(G) smaller than the number of disks(N) increases the load by the declustering ratio denoted by α = (G -1)/(N -1), which can be lesser than that in Raid 5 while switching to, and subsequently operating in rebuild mode. The Nearly Random Permutation (NRP) layout provides the flexibility to vary the declustering ratio (α) for a given N, and the Vacationing Server Model (VSM) of processing the rebuild requests provides acceptable rebuild and user response times.

The rebuild performance and the user response time can be improved by introducing an onboard buffer in the disks, which caches a single track upon arrival of a rebuild request while in rebuild mode. Such an enhancement is proposed, and the architecture is described along with an analysis using the DASim simulation toolkit developed at NJIT.

Also proposed is the delayed termination of vacations with two user requests as this improves the rebuild performance with a negligible negative impact on user response time. Finally, the effect of limiting the rebuild buffer on the rebuild performance is presented in the context of three different disk utilizations and declustering ratios.