Restart services for highly available systems
Nicholas S. Bowen, Christos A. Polyzois, et al.
IPDPS 1995
The response time of disk accesses in RAID5 disk arrays degrades when one of the N + 1 disks fails and there is a further degradation by the interference caused by rebuild processing. In addition to giving user accesses a higher non-preemptive priority over track reads for rebuild, we consider: (i) the read redirection option; (ii) split-seek option, i.e., allowing user requests to preempt track reads after a seek is completed; (iii) the split-latency/transfer option, i.e., allowing user requests to preempt track reads, Simulation results show that for lower disk utilizations the split-latency/transfer option is desirable in that it reduces response time, while not increasing rebuild time considerably. At higher disk utilizations the response time attained by the split-latency transfer option is comparable to that attained by less preemptive options with shorter rebuild times. A new metric, excess cumulative response time, is defined to compare the efficiency of rebuild options.
Nicholas S. Bowen, Christos A. Polyzois, et al.
IPDPS 1995
Navin Budhiraja, Keith Marzullo
IPDPS 1995
P.A. Franaszek, J.R. Haritsa, et al.
ICDCS 1992
Alexander Thomasian, Behzad Nadji
CSSE