Regular (gen1) SSD's Kneecap Linux Performance

[edwinpj]

These drives offer fast reads... as long as you're not writing anything to the disk, it works well.

Random write performance is bad. OK for web browsing, terrible for a development machine. I bought four and thought that with an Adaptec 5805 512MB caching RAID controller in a RAID 0, I'd be able to get the manufacturer-advertised write throughput.

Man, was I wrong. My Linux system (i7 920, 12GB RAM), is rendered unusable during a Linux kernel compile, or even just while copying a large file. I can't move the mouse, and the system hangs for 10-30 seconds at a time.

No amount of tweaking has solved this problem, including the configuration of:

1.) I/O scheduling algorithm (deadline or noop).
2.) I/O queue depth (per-LUN and per-drive).
3.) Read-ahead throttling.
4.) Limit number of adapter control blocks.
5.) Adapter I/O timeout.
6.) Stripe size.
7.) Toggling AHCI.
8.) Various other driver options.

According to Adaptec, this problem is due to very slow response times from the hard disk (Google adaptec answer aacraid scsi hang, and see the first two hits). The drive does not adhere to SATA-II spec response times. If you have a high-end caching controller that expects true SATA-II compliant disks this product will not work correctly.

I have spent $1,000 on disks that were advertised as high-performance that are terrible for doing real work. Certainly with a good controller, I expected performance approaching what was advertised. I get about 1/4 of advertised write speeds, and the system can't do anything that approaches that limit. I'd be better off with one spinning SATA disk. I wish I had been warned by the manufacturer. I would have gladly paid extra to avoid one month of frustration. Now I am looking at another month of delay while I deal with the RMA/upgrade process.

I hope to RMA these drives to G.Skill and replace them with 128GB Falcons (for a price, I'm sure). If G.Skill helps me to upgrade these drives, I will be sure to let everyone know that G.Skill supports their customers 100%. Hopefully, they will respond to my RMA request soon. Otherwise, I will be sure to follow-up here.

[FaNIX]

Um... that may work with cars, trading in your old model, but surely it won't work with Computer Components, and expecting this from ANY manufacturer is quite silly... Get real, sell your old one, and buy a new one, as easy as that, don't make it G.Skills problem.

[SSDriver]

Um... that may work with cars, trading in your old model, but surely it won't work with Computer Components, and expecting this from ANY manufacturer is quite silly... Get real, sell your old one, and buy a new one, as easy as that, don't make it G.Skills problem.

I have to agree. Before spending that much money, you should know what you are buying. There are plenty of review sites out there. SSD technology is changing every day, we all took a risk when we bought in. I feel your pain, but I don't see how it is G.Skill's problem. They spec'd their drives the same as every other vendor of comparable drives. Sell the drives on eBay, get some Falcons.

[edwinpj]

I bought a SATA-II drive.

I expect it to adhere to the SATA-II specification. So does my hard drive controller.

If the drives I purchased, labeled as a SATA-II drives, adhered to the SATA-II specification, I would not be having a problem. Also, I read many reviews for this and similar drives, and nothing lead me to believe that a caching raid controller could not handle them.

The response times of the drive are out of spec, which is why a controller that expects true SATA-II compliant hardware does not work correctly. The controller correctly handles request timeouts as an error, which causes it to flush its cache. It does not keep queuing requests to a non-responsive drive. This is exactly what I would expect well-designed hardware to do. The goal it to avoid data errors. Seriously, don't you think a controller with 512MB cache should be enough to put in front of four drives in a RAID-0? That's 128MB/drive, better than Falcon!

Look, I don't think G.Skill meant ill here, and like I said, I did all I could to get things working right. If you sell a 128GB hard drive for $250 and say it's SATA-II compliant, it better be, because right now, I'd be better off with a $85 1,000GB spindle.

I know that other users have had success with Adaptec aacraid controllers with Falcons, as they use a different controller, so that should be fine.

P.S.-- G.Skill is working with me on this issue. They understand that customer service is important, and they want to win more high-end customers to take the lead in the SSD market.

[Mor]

Hi you did upgrade your Adaptec's BIOS to the latest (16501 18th Feb for a 5405) before I did this I had exactly the same trouble with Falcon's, after upgrading the Adaptec's BIOS all has been good (Falcon firmware 1370). Thus it might be the SSD's or it could be your controller's BIOS buying it recently is no garantee of it having the latest BIOS. Goodluck in solving your issue/s.

[edwinpj]

Yep, updated the firmware.

I had the latest everything. I really tried all I could find. I spent weeks focusing all my energy on getting it figured out. Then, I come to learn from Adaptec that there's not much to be done if the hardware is out of spec. I was even willing to change the driver if I had to (it's open source!!! gotta love Linux...) But that low-level error handling is in firmware, and that's not open-source.

Thanks for your suggestion, though. I did try one of your tips: setting the cache to drive-specific. That basically locked up my system unrecoverably under load! I had to do a hard reset. It's just not a Falcon. If there was a way to make it work, man, I would do it.

I hope that G.Skill can get their Falcon drive certified with Adaptec aacraid controllers. Then, they'd be the apes bananas for sure with high-end users. Right now, the Intel X-25M is the only reasonably-priced SSD that's officially certified by Adaptec, which really should change. I'm sure the Falcon could pass muster in Adaptec's labs. Once they do, they'll have not just a hit: they'll have a mega-hit on their hands.

As of now, G.Skill seems to be working with me to rectify this problem. I'll let y'all know how it goes.

[Mor]

Just to be sure you have the latest from Adaptec, be aware that if you install the latest driver package (8th JAN) it can only install BIOS 16343 and the latest ASM (16th Jan) will not install the latest BIOS. On the page for your Adaptec controller (just checked) top left there is 'BIOS updates and other updates". http://www.adaptec.com.au/en-US/down...ptec+RAID+5805

I am only pushing this clarification because I had virtually given up due to the never ending mouse hangs when the computer was not even really doing any work, thinking I had the latest but did not due to Adaptec's page not being very straight-forward

[edwinpj]

Yep, just triple-checked the firmware in the Storage Manager. 16501. Thank you for your thoroughness. Are you an engineer too?

It really bugs me to have so much invested in something that's not even close to living up to expectations. At least I thought it would be a little faster...