Matthew Roberts wrote: > I wrote: >> I have been experimenting with a softraid mirror, using two cheap SATA >> disks. >> In general the performance is very good - except when rebuilding. A >> quick >> set of sums suggests that the problem is seek time. >> >> The disks are 7200rpm, therefore one can hope for 120 seeks per second. >> "systat iostat" (while rebuilding) gives this (edited to fit in an >> email): >> >> DEVICE READ WRITE RTPS WTPS SEC >> wd0 3801088 3827302 58 59 1.0 >> wd1 0 3801088 0 58 0.0 >> sd0 0 0 0 0 0.0 >> Totals 3801088 7628390 58 117 1.0 > > [snip] > > I've patched src/sys/dev/softraid_raid1.c so that my machine doesn't write > back to the source chunk(s) when rebuilding - making sure that the other > writes go to both disks. > > Now my rebuild speed is much better: > > DEVICE READ WRITE RTPS WTPS SEC > wd0 61525196 9830 938 0 0.5 > wd1 0 61525196 0 938 0.5 > sd0 0 0 0 0 0.0 > Totals 61525196 61535027 938 939 1.0 > > --- > > FWIW: I found that the rebuild code is (currently) only used by raid1, so > there is no benefit in increasing the buffer size for all rebuild vs > tweaking > the rebuild logic. > > --- > > I've tested this patch on a virtual-box machine with a 3 way raid 1 mirror > and it seems to do sensible things there. (i.e. reads from the good disks, > rebuild writes go only to the bad disk, other writes go to all disks and no > data corruption). > > I'm now running it on my (experimental) real machine... and the performance > increase is substantial, as you can see above. (Fingers crossed that my > data will be safe). > > The patch feels a bit "copy-and-paste"y, and I'll tidy it up in a day or > so (and repost). But if you want to risk it, here it is:
When reposting, be sure not to mangle the tabs in the diff again... ;-) [mangled diff stripped]
