System and method for improved rebuild in RAID

A system and method for improved RAID rebuilds under host IO conditions, that greatly improves rebuild times and prevents host IO starvation. A queue in a drive that is part of the RAID is used to store rebuild and host IO requests, with rebuild IOs issued to the head of the drive queue. Rebuild requests in the drive are delayed by a delay time. This delay ensures there is no unintended side effect of this invention that may result in host IO starvation for the RAID system. Delay is computed as a heuristic function from a plurality of variables, such as disk head response times, time allotted for rebuild, number of rebuild threads, drive behavior, rebuild IO profile, host IO profile including workload and locality, and a timeline of sequenced events, which in one preferred embodiment cycles every 70 seconds. The rebuild IO profile has a rebuild rate presented as a plurality of options, e.g., "no impact", "low impact", "high impact" or "max. impact" for how fast the rebuilds will occur vis-a-vis the host IOs and how the rebuilds will impact host IO operations. In certain configurations under heavy host IO conditions, the rebuild times employing the method and apparatus of the present invention can be improved by up to a factor of ten.