Very poor RAID0 performance

[surgeklick]

I've just created a RAID0 array using 2 x Samsung Spinpoint T166 320mb with my GA-P35C-DS3R. I used the ICH9R driver on the motherboard driver CD. The HD are plugged into SATA Port 0 (Master) and Port 1 (Master). I've installed XP and the raid works fine, but running HDTune shows very poor performance.

Average transfer rate = 78.1 MB/s. Min is 70.8, max is 93.0.
Access time is 15.9s.
Burst Rate = 117.6

In the Tom's Hardware RAID Scaling Charts (http://www.tomshardware.co.uk/raid-scaling-charts-uk,review-2340-10.html) using the same model HD's give transfer rates around the 150-155MB/s mark, so around double mine. I know these charts were using a hardware controller, but surely that can't account for much a large performance difference.

Any ideas what is causing such poor performance?

 

[chunkymonster]

Two things come to mind, 1) use the latest chipset drivers from the Gigabyte or Intel website rather than the drivers off the install CD, and 2) onboard RAID is known to be slower than a dedicated RAID controller card as the cpu must calculate parity as well as process for the system.

Not for nothing but the Areca 1220 controller card is one of the best and I'm not surprised that there is a considerable performance difference between it and the ICH9R.

If you update the mobo chipset with the latest drivers from the Gigabyte website as well as the latest RAID drivers from either Intel or Gigabyte, you should notice a performance increase. But, do not expect the same results or performance as having a dedicated RAID controller card.

Good luck!

 

[SomeJoe7777]

You may also want to check your stripe size - there have been some people report that using anything other than the default stripe size of 64KB on the Intel controllers results in a performance decrease.

The Samsungs are not the fastest drives in the world, but the difference between them and Seagate or WD 320GBs should be a few % at most.

 

[surgeklick]

Thanks for the helpful responses.

Would a software controller neccessarily only be half the speed of a hardware one? The performance I'm getting seems to be the speed of one HD on its own, so it would give no RAID0 boost at all.

My stripe size is 128kb at the moment. I might have to wipe the current RAID and create a new one (losing all the drivers etc I've been installing over the last couple of days).

One option I didn't mention is Volume Write Back Cache. At the moment it's set to false, would enabling this make any difference?

 

[surgeklick]

Also, I had a choice of RAID controllers between the Intel ICH9R SATA RAID Driver and GIGABYTE SATA2 RAID Driver. Would the latter have been faster because it (like my drives) are SATA2?

 

[surgeklick]

Sorry to truple post, but this is resolved now. I enabled the Volume Write Back Cache and my transfer rates bumped to around 125mb/s. Still slightly less than I was expecting, but good enough not to spend all day reinstalling everything.

Thanks again.

 

[enlightenment]

Enabling write cache means you can loose portions of your filesystem in case of a crash or power failure -- know that you are running in a particularly vulnerable setup if you leave this enabled!

Second, you are testing with flawed benchmark software. You should use a benchmark which either:
- tests non-blocking on the filesystem
- tests on the raw device using a queue depth higher than 1

I suggest you download and run the SiSoftware Sandra filesystem benchmark (not physical disk) instead, or use the IOmeter advanced benchmark suite (though far less easy to setup). Sandra tests directly on the filesystem, and you gain benefits like multiple I/O queue depth, read-ahead, write buffering and other optimizations your filesystem delivers. This is required to unlock the power of RAID.

Then i suggest you re-test the effect of the write back option in your driver, and consider if you should use it. I can't stress enough that if you leave this option enabled, you should have a fresh and damn good backup at all times or the data you store is unimportant and replaceable. Else you are gambling with your data.

 

[SomeJoe7777]

Enabling write cache means you can loose portions of your filesystem in case of a crash or power failure -- know that you are running in a particularly vulnerable setup if you leave this enabled!

Then i suggest you re-test the effect of the write back option in your driver, and consider if you should use it. I can't stress enough that if you leave this option enabled, you should have a fresh and damn good backup at all times or the data you store is unimportant and replaceable. Else you are gambling with your data.

In his case, the write-back cache can be somewhat risky for the reasons you mentioned, although he can mitigate it somewhat by placing the computer on a UPS. This doesn't fully protect from OS crashes or other problems, but will at least address the most likely source of an interrupted write -- power failure.

Other cards like the LSIs and 3Wares can remove the risk entirely by using a Battery Backup Unit (BBU) on the RAID card. However, the ICH9R does not have such an option.

 

[enlightenment]

ICH9R is driver RAID and uses RAM as buffercache. So a BBU would not make much sense: after rebooting if a crash occurs the data is gone anyway. A hardware controller with BBU can write data to the disks even after a crash, reboot or power failure.

The sensible thing to do is making sure you have a *good* backup at all times, then the lower reliability of the software RAID array does not matter much.

 

[Guest]

Just a quick note, in Windows Vista, there is the Vista Experience index where it rates your processor, RAM, storage system, graphics, etc. Not that the original poster is using Vista or anything, I just thought this was worth sharing.

Going from a single drive to two drives in RAID0, my HDD score went from 5.1 (as I recall) to 5.9. According to Microsoft, the best score possible with present-day tech is 6.0. I think the whole scoring system is slightly suspect, but it's nice to have SOME objective evidence that my rig didn't get SLOWER according to the Vista Index.

I thought this is interesting in that Vista actually has a sort of "benchmark" utility built in. FWIW.

 

[jitpublisher]

The Vista performance index is a really, really poor way to judge your systems performance. It is not a benchmarking tool, it simply takes your current hardware setup and gives it a preset "rating" factor. In other words, the new hardware you just installed could actually be performing worse in real life than the old hardware due a driver issue, or some other problem, and Vista may show improved rating.