The estimated error rate is however about 1 in 100,000 pages, which isn't quite good enough. Probably a slightly better code can be used with a small amount of inflation; even something like RAID-5 except over pages (i.e. parity pages) would do.