Optimizing a highly fault tolerant software RAID for many core systems
Henning Klein, Jörg Keller · 2009
We present a parallel software driver for a RAID architecture to detect and correct corrupted disk blocks in addition to tolerate disk failures. The necessary computations demand parallel execution to avoid the processor being the bottleneck for a RAID with high bandwidth. The driver employs the processing power of multicore and manycore systems. We report on the performance of a prototype implementation on a quadcore processor that indicates linear speedup and promises good scalability on larger machines. We use reordering of I/O orders to ensure balance between CPU and disk load.