Stuart Levy schrieb:
> Simon Matter suggested, re improving setup of a RAID-5:
> > 2) The really important thing is how you set up the raid5 array. Did you
> > make an external log? That makes the BIG difference. I suggest this
> > layout:
> > Every disk has:
> > Part1: 200M Part2: 8.8G
> > create raid1 on disk 1-2 / part 1
> > create swap on disk 3-8 / part1
> > create raid5 on disk 1-8 / part 2
> > create XFS FS on raid5 with external log on raid1
> Just wondering about the importance of a redundant
> external log. I have a RAID-5 setup for the main filesystem
> data, but keep the log on a nonredundant partition of
I have made several performance tests some time ago. I learnt that,
while it's imporant to have external log, it's not so important to keep
the log on different spindles. I like to have everything redundant,
rootfs, datafs, logs and even swap. Swap was dangerous on software raid
long time ago but it seems safe today. Putting the log on raid1 will
improve security and read performance while write performance is almost
identical to a single drive. That's why I always do it this way.
> another drive, on the theory that if the log were lost,
> I shouldn't lose more than the last few seconds' activity.
> And it appears that xfs_repair -l /dev/newpartition -L /dev/maindevice
> can be used to initialize a new external log -- I won't need
> to re-create the entire filesystem or anything.
> (I tested the above create-new-log scenario, though only on a
> cleanly-unmounted filesystem.)
> Also, for performance, I imagine that it'd help a lot to have
> an external log on a *different spindle* than the
> main filesystem, to avoid lots of extra seeks.
> Since N disks fit in my system, I used N-1 for the RAID-5
> main xfs filesystem and the other 1 for the external log
> (and system disk). But if the ext log needed to be *both*
> redundant and to share no spindles with the main fs,
> I'd be limited to N-2 disks for XFS data, a big loss.
> Does this sound right?
> Stuart Levy, slevy@xxxxxxxxxxxxx