Description
This video demonstrates the benefit from separating data and parity on a RAID-organized SSD. We run the same Zipf workload as in the RAID5-Parity demo, in which the size of all write requests is one page. However, the RAID-5+ manager write data and parity pages on different blocks, because the parity pages are “hotter”—they are updated more frequently than the data pages. Storing them in a separate partition results in them occupying less space in the device. The parity update overheads stays o.5 (i.e., one parity update per write), but the write amplification is reduced (compared to the regular RAID5 manager).