Disk array system which performs data reconstruction with dynamic load balancing and user-specified disk array bandwidth for reconstruction operation to maintain predictable degradation

摘要

A disk controller and method which dynamically compensates for host command queue depth during the data rebuild process and thus maintains a predictable level of performance during data reconstruction operations. During non-idle periods, a rebuild task monitors the current host command queue depth generated by the host and submits additional rebuild requests accordingly. Rebuild requests are preferably sized based on the current rebuild queue depth and the user-selected performance allotment for rebuild operations to maintain a predictable level of performance degradation. Therefore, the rebuild task dynamically compensates for host command queue depth by queueing an appropriate number of rebuild requests of varying size so that neither requesting task dominates. This compensates for instances when the host queues requests deeply, such as during write cache flushing, while also preserving responsiveness when the host has one single thread of activity. In an alternate embodiment, the disk controller includes a first queue which is relatively deep and a second intermediate queue which is relatively shallow. Host requests are queued into the first queue and are then filtered into the intermediate queue. The rebuild task queues rebuild requests directly into the intermediate level queue, whose command queue depth is limited to very few requests. In this embodiment, the variation in command queue depth due to host requests is filtered out and by the first queue, and performance degradation is controlled more consistently.