> -----Original Message-----
> From: Mike Castle [mailto:[EMAIL PROTECTED]]
>
> Man, that sounds like a bad personals ad.
>
> Anyway, just getting into RAID. Non-critical home system
> that I just want
> to play around on. It's an older 233MHz based system running IDE. In
> otherwards, I'm not too concerned about performance. Though
> if I could
> arrange things to not hurt performance, and perhaps even
> increase things,
> that would be great.
Heh, that's a lot faster than my primary RAID testing system. I run a P166
with 128MB of ram, and an HP AHA-2920 (with bios) to test. It's not fast,
but it's rock solid stable.
> Running 2.2.17 with software RAID patches.
Are you using the new RAID tools?
> Running 2 built in IDE controllers (Intel 430VX chipset, not
> that great,
> but free), and an old AWE32 as a 3rd controller. As I said,
> I'm not in it
> for the speed. On the other hand, I have these as my harddrives:
>
> hda: Maxtor 91303D6, 12427MB w/512kB Cache, CHS=12624/32/63, (U)DMA
> hdb: Maxtor 53073U6, 29311MB w/2048kB Cache, CHS=59554/16/63, (U)DMA
> hdc: Maxtor 93652U8, 34837MB w/2048kB Cache, CHS=4441/255/63, (U)DMA
> hdd: Maxtor 92048U8, 19470MB w/2048kB Cache, CHS=39560/16/63, (U)DMA
> hdh: Maxtor 53073H6, 29311MB w/2048kB Cache, CHS=59554/16/63
UDMA drives on PIO mode 3 controllers? Is that last drive really not UDMA?
> Yeah. About 120G or so.
That's a fair chunk of disk...
> Anyway, I want to parition them into something like this:
>
> / 128M
> /var 64M
> /tmp 128M
> /usr/tmp 128M
> /var/tmp 128M (yeah, figured might as well keep all three
> separate)
> /home 10G
> /usr 10G
> /var/spool/news 25G (leafnode mirror, don'tcha know)
> /usr/src 15G
> /usr/mirror 40G
> /usr/music 13G
Looks like 11 slices, not including swap.
> Plus probably 64M of swap on each physical drive.
So you want 320MB of swap? You should probably just put your fastest 3
drives as masters, and have 1 swap partition on each.
> Numbers don't quite add up. Probably extra to mirror.
Mirroring cuts your space in half, obviously
> Anyway, obviously I can't follow the recommended procedure of no slave
> drives. So I'm going to face IDE contention. Probably best
> I can do is
> reduce head movements.
Avoiding slaves is not only for speed, but also for reliability. Often when
a master drive is lost, the entire IDE channel goes down.
> I'm considering keeping /, /var, and {/usr,/var,}/tmp as non RAID.
> Probably one per disk just to spread things out. Then taking
> the rest of
> the disk space and using it with RAID.
O.K.
> The biggest question is, would it make more sense to spread
> everything out
> across all 5 drives, or to go no more than 2-3 drives for a particular
> partition? If no more than 2-3 drives, would it make sense
> to put say,
> both as masters, or have 1 master, 1 slave for a particular
> md partition?
Go with a max of 3 drives, 1 partition on each, for each md device. I'd put
non-critical stuff onto your slave drives (which I would also make the
slowest drives). That will give you the best reliability and speed, and
since this system isn't all that fast to begin with, maximizing speed is a
good thing.
> One thing I was considering was just dividing the disk into arbitrary
> chunks of 1-2G. Then just piecing them together. Of course,
> partitions on
> the same drive would be linear, then across drives would be
> raid-0. I was
I'm not sure how the RAID code handles that. I can't think of any reason
that I'd want to chop things up like that, especially considering how much
space you've got available. At 2GB chunks, that's about 60 chunks. Yikes.
> considering doing it like this so that, if necessary, I could reassign
> slabs and moves things around as necessary (after doing things like
> resize2fs and mkraid). Would this be feasible (running out
> of md devices)?
I don't know. Linux uses the traditional DOS partition model by default, so
that gives you a max of 7 useable slices per drive. I generally tend to
keep as few different slices on a disk as I possibly can. My single disk or
mirrored machines have 3 slices per drive. /boot, swap, and /, in that
order. Machines with more hard disks gets a bit more complex, and it really
depends on the application.
> Hurt performance (going through both raid0 and linear)? Too much like
> micro managing and it's just not worth it?
I don't think you can layer RAID levels using the 2.2.x kernels. The only
levels that will even work are RAID 1 combined with RAID 0, and those have a
fatal bug where it doesn't recognize the failure of a disk, so that defeats
the purpose of RAID 1. If you haven't already done so, you should read the
Multi-Disk System Tuning HOWTO. It's available from
http://www.LinuxDoc.org/ and it's mirrors. HTH,
Greg
-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to [EMAIL PROTECTED]
