Basic RAID question

[mroz]

It seems to me that IMR, or practically any other software RAID solution for that matter, is ideally suited to those seeking performance and redundancy on a budget. Like partitions, IMR will suffer when accessing data on both parts at the same time.

Does anyone have performance stats comparing the following:

a) Copying between two drives

b) Setting up two RAID 0 using a pair of drives & copying from one RAID to the other?

That'd be useful to give an idea of the worst case performance loss as a result of configuring the drives in this way.

I'm considering RAIDing the drives in my next system & in the past have used a pair of drives, one for OS & scratch space, one for data. Thus I'm often doing disk intensive stuff involving shifting data between drives.

I'm wondering now, if I go with a pair of drives configured to give me a RAID 0 for pagefile & scratch and a RAID 1 for OS & data, what performance loss I'll see for cross RAID transfers. Can't seem to find this info anywhere.

FWIW If I go with this option I'll later add another pair of drives with a further RAID 0/RAID 1 split for scratch / data back ups, which will eliminate most of those losses.

[EtheAv8r]

I saw somewhere they did a comparison of a separate Raid 0 and Raid 1with the IM Raid 0/Raid1 and the performance difference was pretty much negligible.

http://techreport.com/reviews/2005q1...d/index.x?pg=1

have a look through there, they basically conclude that there is no downside to using IM raid. The only problem you will have is if you try accessing data from both Raid 0 and 1 on the IM arrays at the same time, as the disk heads will have to access both at the same time, so copying data from raid 0 to 1 or vice versa etc, but generally I think the benefits would outweigh the problems. (mentioned here http://www.techarp.com/showarticle.a...tno=206&pgno=1 )

This is usefull, but I would comment that these reports are over 2 years old, and the later marks of the Intel ICHnR controller have got faster and the software better. Also in the first artical they created the RAID 1 first, and the RAID 0 second. The objective of RAID 0 is speed/performance, and they lost a lot of this potential by creating the RAID 0 second, and therefore on the slower inner cylinders. Had they created the RAID 0 first, there would have been a marked improvement methinks.

[iranu]

You may also wish to have a quick read of this if you haven't already. http://forums.hexus.net/showthread.php?t=115697

Funnily enough, whilst doing abit of research I came across the intel raid matrix thingy yesterday.

[mroz]

Also in the first artical they created the RAID 1 first, and the RAID 0 second. The objective of RAID 0 is speed/performance, and they lost a lot of this potential by creating the RAID 0 second, and therefore on the slower inner cylinders. Had they created the RAID 0 first, there would have been a marked improvement methinks.

According to the second page they used two configs for testing IMRAID; the first had RAID 1 first which they used to provide the RAID 1 stats only; the second had RAID 0 first & again was used to supply only the RAID 0 stats.

[EtheAv8r]

According to the second page they used two configs for testing IMRAID; the first had RAID 1 first which they used to provide the RAID 1 stats only; the second had RAID 0 first & again was used to supply only the RAID 0 stats.

oppss that'll teach me to quick read!... I stand korrected.