X25's in raid. What is the optimum stripe size?

**Ao1** · 02-22-2009, 12:00 PM

I'm using two X25e's in raid 0 on an adaptec 5405 card. The strip is set at 128. Has anyone used different strip size and if so what read/ writes are you getting?

This is what I get with 128 strip.

**dengyong** · 02-22-2009, 12:39 PM

128 should be fine.

**One_Hertz** · 02-22-2009, 02:05 PM

1. Calmly take your adaptec 5405 out of your computer.
2. Sell it.
3. Hook up your X25-Es to onboard raid.
4. You've just doubled your small read random and sequential IOPs.

The raid card you have slows down your SSDs A LOT. I have the exact same setup as you (5405 + two X25-Es) and have done a lot of testing. Onboard is faster in almost every way. This is of course considering your onboard isn't gimped somehow.

Strip size of 128 is fine.

**Ao1** · 02-23-2009, 02:20 AM

Hi One_Hertz,

Using the Atto benchmark (for what it's worth) the X25e's are working much faster on the 5405, although it's not a tangible increase in terms of user experience.

The reason for going to the 5405 was two fold. First I have had nothing but problems with onboard raid using Vista 64 despite changing from Nvidia to Intel. If nothing else the 5405 is stable and I haven't had a single problem with it. (Raid failures)

The second reason was that I felt sure that the X25e's were throttled by the onboard controller, which seems to to be the case if you can believe the Atto results.

I'm not a "benchie". I'm only interested in stable and fast user experience. Can you explain point 4 in your post? My guess was the card interaction with these drives could be optimised with the strip size.

Thanks

**One_Hertz** · 02-23-2009, 08:23 AM

There is something wrong with your onboard controller it seems. The adaptec card just about DOUBLES the latency of the X25-Es. It reduces their small random and sequential reads by about 50%. It is a big bottleneck for those SSDs. My ICH9R is working perfectly fine, 500MB/s reads 390MB/s writes in Atto.

**Knight** · 02-23-2009, 09:08 AM

Enable volume write back cache for the Intel controller. I have two drives in raid0. With it disabled I was only getting the performance of a single drive. (~85 MB/s) With it enabled I was seeing about 160MB/s.

**One_Hertz** · 02-23-2009, 09:24 AM

Originally Posted by Knight

Enable volume write back cache for the Intel controller. I have two drives in raid0. With it disabled I was only getting the performance of a single drive. (~85 MB/s) With it enabled I was seeing about 160MB/s.

You werent getting the performance of a single drive. Your benchmarks (I guess hdtach/hdtune) showed you numbers of a single drive, but they were wrong.

**Ao1** · 02-23-2009, 10:44 AM

Originally Posted by One_Hertz

There is something wrong with your onboard controller it seems. The adaptec card just about DOUBLES the latency of the X25-Es. It reduces their small random and sequential reads by about 50%. It is a big bottleneck for those SSDs. My ICH9R is working perfectly fine, 500MB/s reads 390MB/s writes in Atto.

Hi One_Hertz,

From my experience read/ write benchmarks are not indicative of how a SSD drive will perform in real usage.

I guess that X25e's are completely overkill for a home rig, but I wanted to switch over to SSD and having tried the OCZ core it seemed that you had to let go of the cash if wanted a SSD to perform any better than a laptop HDD. (That statement is based on user experience. Advertised read/ write speeds are nothing to go by in terms of real time performance if the OCZ core is anything to go by. The drive just locks up when under any pressure and overall you get the experience of using a slow laptop HHD.)

Having spent the cash for the X25e's I'm really happy with them, but I would like to know if they are set up to the run at their fastest and the strip size seemed to be the only way possible of tweaking them.

It seems a bit ironic that SSD are really good at large file reads, when having large files is not an option due to the small size of the drives

I'm much more interested in small read/ write performance and high io's.

Going back to onboard controllers is not something I want to do as the 5405 is so much more stable and seems to help overall PC performance. Have you tried smaller or higher strip sizes? Did it make any difference?

I curious about why you think that the 5405 will reduce small random and sequential reads by about 50% and double latency. Is this something you could explain in a bit more detail?

Thanks in advance.

**One_Hertz** · 02-23-2009, 11:30 AM

Originally Posted by audienceofone

I curious about why you think that the 5405 will reduce small random and sequential reads by about 50% and double latency. Is this something you could explain in a bit more detail?

Thanks in advance.

It's not that I think, it's that I have done testing (as well as many others) and I know

.

Here is a quick example from Napalm:
http://www.xtremesystems.org/forums/...&postcount=367

He is testing the adaptec 5805 which is essentially the same thing. Access times 0.21/0.31 versus 0.15/0.15 on a different raid card, or using onboard. The difference is more drastic in the case of X25-Es. Onboard my latency is 0.12ms versus with the Adaptec 5405 I had 0.22ms or so (same as what you have now).

Small reads/writes are A LOT faster using the onboard controller (considering your onboard is working properly) and so are the random IOPS.

Adaptec does not officially support SSDs and it shows why.

If you do a quick HDTune random access read test you'll get around 6600 I/Os on the 512byte level. If you stick it on onboard you should get 11500. Those were my numbers anyhow.

**Ao1** · 02-23-2009, 03:10 PM

Hi One_Hertz,

Thanks for the heads up

Seems that SDD drives are still a round peg in a square hole and it doesn't look like that is going to change anytime soon.

I'll quiz Adaptec, but as you say they don't support SDD so I can't see that going very far. Ask intel about strip size etc and they tell you to go to Adaptec.

Still, it's not like the X25e is not performing more than adequetly so maybe I should just enjoy what I've got and not bother to try and find a way of squeezing out any more.

**Ao1** · 02-25-2009, 06:28 AM

OK, so it seems that HDD benchmarks for SSD drives may not be accurate, however the results below are interesting following a number of Win 7 updates released today. I've checked the release info on the updates and it's a bit sketchy on what the updates are for and one update is not even listed (kb962921), however it has had a big difference to the benchmark results as below, although random access time remains the same.

**Ao1** · 02-25-2009, 10:47 AM

Interesting...just tried Hyperfast on the X25E, which has the OS on it and has been running for a couple of months. It shows little or no defragmentation on the drive.