On 2025-03-16, Jon Higgs <j...@altos.au> wrote: > On 16/03/25 11:51, Rub??n Llorente wrote: >> As far as I understand, raid 5 performance is usually slower than you >> would expect because parity data needs to be computed and written to >> the drives too.
softraid's implementation of RAID5 is not particularly quick. (maybe naive but I suspect using cpu vector instructions for the xor operations would probably help). Other issues (lack of boot support, writes fail rather than retry if a member drive has just gone offline, max RAID5 not RAID6 especially combined with the slowness would make me worry about getting drives rebuilt before a second drive fails) would definitely push me towards using something other than softraid for anything more than RAID1. > I just tested this on my machine with a pair of SSDs in RAID1. It was > still under 50MB/s. > > $ dd if=/dev/zero of=./blah bs=1M count=4096 4294967296 bytes > transferred in 91.310 secs (47037206 bytes/sec) > > I don't think it's parity, it seems like it's just slow. Not sure where the problem is, but at least RAID1/crypto softraid is not usually that slow. The only place I can think of where I have softraid RAID1 at the moment is on 2010-era hw (2TB Hitachi Deskstar 7K2000, HP Microserver N40L) and even there I get better speeds than you're seeing: $ dd if=/dev/zero of=./blah bs=1M count=4096 4096+0 records in 4096+0 records out 4294967296 bytes transferred in 71.779 secs (59835986 bytes/sec) fwiw, with softraid crypto (no RAID) I get this on a more recent laptop (512GB Samsung OEM nvme, i5-1245U): $ dd if=/dev/zero of=./blah bs=1M count=4096 4096+0 records in 4096+0 records out 4294967296 bytes transferred in 16.046 secs (267665916 bytes/sec) -- Please keep replies on the mailing list.
