[Nottingham] via kt4 Ultra Raid Board...any good with linux?

[Matthew Sackman]

On Wed, Nov 27, 2002 at 02:23:32PM +0000, Lee wrote:
> as for mirroring in software, I've always been a bit dubious about it,
> because NT's mirror is so so bad in software , but maybe linux is a
> breath of fresh air, I know it is in other departments  :-).

Unfortunately, in software RAID, linux sucks big time.

> Can you mirror a boot device in linux with software?

Nope. You can get / mirrored, but not /boot IIRC.

> or is it best just to have three driver's and mirror two of them whilst
> the remaining two data drives are a mirror?
> 
> Murphy's law will dictate the the unmirror drive will fail first..
> :-(((((

Just mirror as much as you want. You can always run a cron job to keep a
copy of /boot on another partition of another disk and make sure that's
bootable.

> so,
> 
> IDE Disk A------
>                |---Primary Linux mirror
> IDE Disk B------ 
> 
> by using a mirror boot partition, do I run the risk of making the system
> hard to recove (fsck/restore from backup) in case of a failure?

Can't do it. Besided, if a disk does go, the BIOS should ignore it and
look at the next device and try and boot off that.

> /rant
> Arrgg, bleeding harddrives, they make them bigger, faster, and cheaper
> but they don't make them more reliable, why doesn't some bright spark
> come up with a ide raid adapter that is totaly transparent to the
> operating system, so to the os it look's like a nomral ide device...easy
> huh..

There are. Some of the high end promise fast trak controllers basically
do this. Personally I would avoid IDE in a mission critical environment
- SCSI is much better for this in terms of reliability and build
quality. You still pay through the nose though which is a pain. Also,
Adaptec, Mylex and others do do some very nice hardware RAID SCSI cards
which are pretty much transparent.

> I used to have a xyratek scsi raid system that did the same thing...
> it was very sexy, if hardware can be sexy

It can be. I have a zaurus. ;)

> as for naan, you can check the web site if u want, you'll all be please
> to know that the all of the network infrastructure will be linux
> based.....:-)))).... somebodies come up with linux drivers for the 22mbs
> cards too (actual throughput around 8) , so even more fun to be had me
> thinks..... I keep getting offer of help, and generous donation of
> hardware and software , it really has inspired me to get this thing off
> the ground and everyday it seems like much more of relaity instead of a
> pipe dream.

Sounds good.

Linux software RAID sucks because it is massively inconsistent. I had to
do a lot of work with this stuff over the summer. I had two machines,
and one was primary server (P) and the other was a hot spare (S). Using
ENBD, S exported a file system to P which then mirrored it with a local
partition. The idea was that P or S could go down (but not both), and
the contents of the filesystem would *always* be available. It used
heartbeat which is excellant.

The problem came when if P or S went down and then came back up the
system had to know which way to sync the RAID array in order to maintain
(and not over write) the filesystem. This was the stumbling block
because the software RAID tries to be intelligent and stores information
about the status of the array on the partitions themselves. So if I left
it to itself it would either resync to the wrong disk or it would just
give up. It was a right PITA. So I ended up having to use a whole bunch
of arguements to force md to behave. I was a real pain, and the guy who
wrote/writes enbd agrees that both software RAID and LVM(1) is complete
poo.

Anyway, the whole thing's documented (badly) on
www.wellquite.org/redundant/ if you want to read it. Scripts are there
too.

Oh and London is fine as you were asking. It's nosey and smelly and
tiring and they're trying to teach me Java now, which as they spent the
first 6 weeks teaching me Haskell is a bit of a let down - Haskell is
fantastic, and Java is distinctly not in comparison. Still I've got 4
years to get used to it... ;)

Matthew
-- 
Matthew Sackman

BOFH excuse #287:
Telecommunications is downshifting.