Thoughts on software RAID

[btisdall]

Hi folks,

tricky subject (and I know the best practice recommendation is for RAID 0+1) but is software RAID 1 considered ok for Scalix?

How does software RAID 1 stack up against a single disk?

Is there much of a performance gain to be had by using one mirrored set for the OS & another for Scalix?

I'm looking at a 3GHz xeon with 2GB RAM & SCSI disk subsystem, the box is only running Scalix, 65 users & very lightly loaded.

Any thoughts on this greatly appreciated.

Ben.

[dahido]

If I were you, based on the user load and server specs, I would go with the RAID 1 just because performance shouldn't make a difference and having the reliability of the RAID can be extremely valuable.

Jay.

[ng]

hi ben,

we have a customer who has had a system with software raid and as long as scalix 10 was running everything was fine, but after upgrade to 11 system performance changed.

we tried everything to keep this server as it is but at the end we changed to hardware raid because disk i/o was the main reason for bad performance. during peak time users had to wait more then 10 seconds when sending an email in outlook without smart cache.

so for a small group of users aprox. 25 i would say its ok but everything else better don't try.

hope this helps a bit.

nikolaus

[grubi]

Hi folks,

tricky subject (and I know the best practice recommendation is for RAID 0+1) but is software RAID 1 considered ok for Scalix?

From my experience I'm not a friend of software RAID. Try to use a RAID controller instead. They are not to expensive nowadays.

How does software RAID 1 stack up against a single disk?

Write operations are slower
Read operations are at least of same speed

Is there much of a performance gain to be had by using one mirrored set for the OS & another for Scalix?

I would not expect a performance gain from this setup

grubi.

[mito]

A linux software raid1 is actually slower than a single disk, because the OS has to write the data twice, pushing the data through the entire system, twice. A hardware raid in this same sense is a LOT faster, because the OS and system only has to write the data once, and then the controller doubles the data to it's two (or more) drives. This lets the system I/O stay cleared up, and available for other things.

I was running a system with a software RAID because the hardware RAID that I had wasn't supported yet. The performance was terribly slow, and I was getting complaints left and right. As soon as support for my Hardware raid was available, I changed to that, and the difference was night and day. *EVERY* aspect of Scalix was faster, even things you wouldn't think relied heavily on I/O.

Also, I personally dislike software raid in general because it's a raid at the partition level, not the disk level. If something were to happen to a disk, you still have to make changes to your /boot partition etc before it will boot properly again. With a hardware raid, all that config is done at the hardware level, and your system wont know that there is any change at all.

One thing to keep in mind though, make sure that the hardware raid that you get has a decent monitor for linux. I've had an LSI-Logic raid card before that had drivers for linux, but no monitoring software, so it was basically pointless. It kept my system RAID'd, but I had no way of knowing if the RAID was still healthy or not (no way to know, other than rebooting and watching the bootup, if one of the drives were having a problem).

Just my thoughts.

[grubi]

A hardware raid in this same sense is a LOT faster, because the OS and system only has to write the data once, and then the controller doubles the data to it's two (or more) drives. This lets the system I/O stay cleared up, and available for other things.

Be careful this is not necessarily true.
A reliable RAID controller (e.g. the whole Smart Array Series of HP does this) will disable the internal write cache of the harddisks. This is the only way for the controller to ensure that the drives are *allways* in a consitent state. This (depending on the kind of write opeation) dramatically decreases the write speed compared to a single disk. To nearly get the speed of a single disk you could enhance those controllers with a battery backed write cache. Reading is faster in most cases as the data could be read interleaved from the two disks. If you want additional speed configure a RAID 1+0 with four disks (A stripe of two disk which is mirrored).

Regards
grubi.

[btisdall]

Thanks to all who contributed to this thread, useful stuff.

Regards,

Ben.

[pharoh]

We deploy raid1 machines and raid5 machines with software raid and boot them via usb keys so we never have to mess with a boot partition on the HDD. works great and fast too. All linux 2.6 software raid combined with lvm2 the linux software raid suite is FANTASTIC we on our larger boxes can hotswap drives and get all the benefits of using the host instead of a little toy board controlling the data. We have found that by using software raid we are not Dependant on a particular hardware vendor and can develop all our tools for monitoring and maintenance and NOT have to worry about what the manufacturer is going to allow us to do differently from another manufacturer. just plain freedom baby.

[racmar]

We actually use a hardware raid 1 solution. I find the convenience to the hard drives to be extremely valuable when hot swaps are needed. I use the Araid 2000 from accordance systems. I can just walk up to the machine and pull out a hard drive, then replace it with a spare. The Araid will rebuild the new hard drive as a mirrored copy of the original without any downtime.

I usually hot swap the secondary drive once a week to carry off-site for a backup.