Raid1 does not increase performance, it will decrease it a little. That's
what your results show: md0 is slower than the slowest of the mirrored
disks, pretty normal. Even though you set the mirrors on two different
channels of the IDE, which is the right thing to do.
You are using an IDE controller (and drives), maybe yours is not very
performant either, because you get a 23% [(2.75-2.11)/2.75] performance
hit, which is probably on the high side. It also seems that your drives
are not identical, which complicates matters a little for the controller.
Cache reads are always much faster than real disk access, until you try
sizes above the cache (memory) size -- then measured 'performance' drops
very fast.
You could try your setup with a newer MB/CPU, it should be slightly better
(maybe 5% hit, ie. similar to your current 6.2% hit on cached access).
TP
On Wed, 19 May 1999, Joachim Zobel wrote:
> I`m running raidtools-0.41 on a P75. I'm using raid1 because I know there
> will be a day when one of my disks will fail. There are three ide disks and
> a cdrom.
>
> Running hdparm on md0 and its components gives me:
>
> dilbert:/root # hdparm -tT /dev/hdd1 /dev/hda6 /dev/md0
>
> /dev/hdd1:
> Timing buffer-cache reads: 64 MB in 3.76 seconds =17.02 MB/sec
> Timing buffered disk reads: 32 MB in 11.64 seconds = 2.75 MB/sec
>
> /dev/hda6:
> Timing buffer-cache reads: 64 MB in 3.78 seconds =16.93 MB/sec
> Timing buffered disk reads: 32 MB in 10.55 seconds = 3.03 MB/sec
>
> /dev/md0:
> Timing buffer-cache reads: 64 MB in 4.03 seconds =15.88 MB/sec
> Timing buffered disk reads: 32 MB in 15.11 seconds = 2.12 MB/sec
>
> Why is this?
> Any hints?
>
> Thanx,
> Joachim
>
> --
> "... ein Geschlecht erfinderischer Zwerge, die fuer alles gemietet werden
> koennen." - Bertolt Brecht - Leben des Galilei
