$2,000 worth (of headache so far..)

Ok team,

 

I've been using the first gen OCZ (ack!) SSD's for a few months with some success, but after much research on the matter settled on my "real" upgrade to 4x Corsair P128's in RAID 0 running on an Adaptec 5805.

 

At first I built it as a secondary array off of a windows 7 install and ran some initial benchmarks. Everything looked peachy. So, I took the plunge and attempted to make it my primary OS target. Install went smooth.

 

It's only been 24 hours and I'm already seeing tremendous signs of stuttering. I'm sure its a configuration issue (me hopes), but I'm 12 hours into my 2nd reinstall and I can't get rid of the stutter.

 

Where I am now, is a clean Win 7 install. Updated the drivers on the Adaptec to the latest. I've disabled indexing.

 

When it runs - it runs GREAT ~1000Mb/sec when looking at 512k blocks. But as you can see I hit < 1Mb/sec and the whole system seems to hang.

 

h e l p

So, just in case it's relevant. here is a little more information on the RAID configuration:

 

All Drives showing up as 3 Gb/s transfer speed

Stripe Size: 512k

Read-cache mode: enabled

Write-cache mode: enabled (write-back)

ROM: 5.2 16343

Device Driver: 5.2 17380

 

Also attaching the 256k block size test.. same outcome...

I thought I had read in some of the tests/reviews posted online, it was better to keep the stripe size at 128k, others have said 64k. I just got the Adaptec 5445, so I'm hoping you find the best solution and post it here. :)

That would change the overall throughput, but not the drop to 1 Mb/sec. My understanding is block size selection should be based on anticipated file sizes, but I could be wrong. I less concerned about maximizing the performance at this point and just want a nice steady throughput with sudden lags.

I've done a little testing with my 3 P128's on an Adaptec 5445 controller and found that a stripe size of 128 offers the best performance overall. Using the 256k block size in HD Tune, I got the attached results. I've also attached the CrystalDiskMark and HD Tach (which I don't really consider accurate)

 

I formatted with the default (4k?) allocation.

 

I hope you find this useful...

Thanks for the update. I'll retool the array with 128 stripe size and the default OS block size. Do you have any stuttering?

 

Dan

I've them all with different stripe sizes and different block sizes. A stripe size of 128 seems to be the sweet spot for the RAID controller, and using different block sizes in the tests 256k, 512k, and 1025k seem to provide pretty decent results in everything except ATTO, which gets a little wierd and results are far outside the averages of the other benchmarks. I'm working on that. I'd love to hear any suggestions you may have on that..

 

I have to say that I'm beginning to reach a conclusion that these SSD drives are much better performing, on a regular user level, as single drives than they are in RAID. As single drives they perform flawlessly and absolutely smoke anything else out there. While they present excellant benchmark numbers in a RAID configuration, they're a little finicky. Manufacturers of RAID cards have had decades to perfect their cards operations with HDDs, I don't think they quite have a handle on SSDs yet. If the user just want's a larger disk, then a spanned volume actually provides very good performance, compared to HHDs in RAID 0, without the quirks associated with RAID 0.

 

I'm not a professional tester, but these are my observations at the moment, from a perspective of 30+ years experience in IT and as an enthusiast. As RAID cards and onboard controllers become more SSD aware, I'm sure my views will evolve. Like anything else that goes fast, there's always more to the overall considerations than speed alone.

 

Here are the ATTO results... and they're virtually identical using a 128k or 256k strip size with the Adaptec controller.

Thanks for the update. I'll retool the array with 128 stripe size and the default OS block size. Do you have any stuttering?

 

Dan

 

I often get one stutter at the very beginning of a test, but that holds true for HDDs also. I figure the benchmark gods just hate me.

check the cpu usage and turn on Kernel times (task manager performance > View> Show kernel times) you may find that your system is been maxed out under kernel use (under norm use there should be little kernel use) nor related to an driver or system bottle neck (that could happen if there is a lot of SSDs is an system)you norm see it when the disks are been maxed out or some times even Light load (had to reload an acer system as the system was stuttering like mad and that was an HDD i blame the speedtouch modem in that case)

 

maxed out kernel times are

dual core its 50% red or more

for quad core its 25% Red or more (i7 with HT on around 12% or more)

 

if any one has tools that they can link me to to diagnose that type of problem with kernel times that be useful save me having to reload windows

We did some testing with 128K Block size but I cant post the results yet, we don't have all of the results yet. But I think 128K would be best with 4 drives and 64K Block Size with two but like I said don't hold me to that till all the tests are in.

 

Are you now able to post the results? I'm going to have to rebuild the array anyway, so nows the time :) Anyone else have any results to share?

 

IRL these drives are SCREAMING fast with this RAID card. There is NO visible stuttering at all -so I take it "back" -- no headache, just inconsistent artificial test results. Real world testing is in, and these bad boys rock.

What raid card will you be using?

And normally the default size will be about the best; but 128K cluster size with 32K for the format was about the best for two M-64 drives in a RAID but there was almost no difference with P-64 drives.

 

Adaptec 5805 -- it's quite nice. I just flashed it with the latest firmware and will be rebuilding it tonight. Will keep you all posted.

Team,

 

I;ve spent a few HOURS with Adaptec Sr. Support. They are **awesome** by the way. They want to see this work. Their most senior tech looked at a serious amount of logging I generated and came to the conclusion that the logs were the "cleanest" he has ever seen. Great, but that doesnt solve the performance problem.

 

I'm not upset at all - the drives are amazing, but I'd like to work with someone to figure out why performance drops off at larger sizes. I'm uploading ATTO benchmarks and the complete logs for my system.

 

RAMGUY *(or whomever) what's the next step to work on this? it's in everyones best interest to see what we need to do to make Win 7 - Adaptecs Blazing Fast 5805 - and 4 of your studly P128's sizzle.

Links to ATTO Benchmarks:

 

http://docs.google.com/Doc?docid=0AaUfT2cNDg5sZGhzYmRuMmdfNmY1azN0cWc5&hl=en

http://docs.google.com/Doc?docid=0AaUfT2cNDg5sZGhzYmRuMmdfOGNwZ3F0cGZk&hl=en

 

Detailed System Specs:

http://docs.google.com/fileview?id=0B6UfT2cNDg5sZWQ4MWFjMTUtZThhNy00NzU5LTkyZTItNGFhZTU0MGJlMzFi&hl=en

 

 

 

Link to support logs:

Logs Files (again, the Adaptec Support said these were the cleanest logs she'd ever seen)

Support.zip

After a TON of reading I settled on 128k stripe and default allocation size. I have tried stripe sizes of 256 and 512 (see earlier in the thread). I can't remember what the default Win7 allocation size is since I've been horsing around with things for a month. :)

 

 

 

Logical device..................0

Logical device name.............PHOTON2

RAID level......................0

Data space......................476.549 GB

Parity space....................0 MB

Date created....................08/25/2009

Stripe size.....................128K

Interface type..................Serial ATA

State...........................Optimal

Read-cache mode.................Enabled

Write-cache mode................Enabled (write-back)

Write-cache setting.............Enabled (write-back)

Partitioned.....................Yes

Protected by hot spare..........No

Bootable........................Yes

Bad stripes.....................No

Power Status....................Disabled

Power State.....................Active

Reduce RPM timer................Never

Power off timer.................Never

Verify timer....................Never

Segment 0.......................Present: controller 1, connector 0, device 0, S/N 09300579-9810366

Segment 1.......................Present: controller 1, connector 0, device 1, S/N 09300579-9810395

Segment 2.......................Present: controller 1, connector 0, device 2, S/N 09300579-9810397

Segment 3.......................Present: controller 1, connector 1, device 0, S/N 09300579-9810386

 

....

 

 

Type............................Disk drive

Vendor..........................Unknown

Model...........................CORSAIR CMFSSD-1

Serial Number...................09300579-9810366

Firmware level..................VBM18C1Q

Reported channel................0

Reported SCSI device ID.........0

Interface type..................Serial ATA

Size............................119.242 GB

Negotiated transfer speed.......3.00 Gb/s

State...........................Optimal

S.M.A.R.T. error................No

Write-cache mode................Write back

Hardware errors.................0

Medium errors...................0

Parity errors...................0

Link failures...................0

Aborted commands................0

S.M.A.R.T. warnings.............0

Phy 0...........................3.00 Gb/s

Power State.....................Full rpm

Supported power states..........Full rpm, Powered off

I hit a good patch and got sick (in a good way) benchmarks.. now do to get rid of the large file stutter...

I have not ever seen a problem with the Samsung based drives with shuddering or stutters, I would suspect there is some other issue from what you have posted. What controller are you using and do you have all four drives setup as one Raid array?

 

The setup is the same as described above (I know you are busy, so I'll repost it here)

 

Adaptec 5805 (Most current BIOS)

4 P128 in a RAID 0 Stripe

128k Stripe Size

Windows 7 RTM - Formatted allocation as "Default"

 

Again, if you can take a look at the earlier screen shots in the thread. Performance SIZZLES up to larger block sizes, then peters out.

 

Dan

Ok team,

 

I've been using the first gen ******** (ack!) SSD's for a few months with some success, but after much research on the matter settled on my "real" upgrade to 4x Corsair P128's in RAID 0 running on an Adaptec 5805.

 

At first I built it as a secondary array off of a windows 7 install and ran some initial benchmarks. Everything looked peachy. So, I took the plunge and attempted to make it my primary OS target. Install went smooth.

 

It's only been 24 hours and I'm already seeing tremendous signs of stuttering. I'm sure its a configuration issue (me hopes), but I'm 12 hours into my 2nd reinstall and I can't get rid of the stutter.

 

Where I am now, is a clean Win 7 install. Updated the drivers on the Adaptec to the latest. I've disabled indexing.

 

When it runs - it runs GREAT ~1000Mb/sec when looking at 512k blocks. But as you can see I hit < 1Mb/sec and the whole system seems to hang.

 

h e l p

 

 

RG -

 

I already did before I had the problems. When it was NOT the system disk all was well. When I rebuilt it as the system drive It went to pot.

 

Some other changes - When i first built the array (same RAID 0, Adaptec,) it was Win7Beta not Win7RTM. It was also an earlier BIOS (I upgraded it to see if that resolved the problem.)

 

Besides that, the only difference was adding a "slow" 1.5 TB media drive to the adaptec controller on it's own channel. I'll TRY removing that.

 

Any other thoughts?

 

Dan