Warning; Possible SandForce 2 SF1200 performance collapse

Hi,
I have just purchased a Callisto 120GB Sandforce 2 drive. While some may never notice an issue I have come across one with my drive. The drive collapses with heavy randomized data files. If the specific firmware doesn't support this you then can speculate it could also result in a bricked drive if all a sudden the compression could not keep up with the data stream.

The problem lies in large writes of even sequential files that are highly further uncompressible (randomized). This could be large pictures, music or even video files etc. The SF2 is known to have a performance issue with these files from the get go but this is different and much worst.

As you can see 4KQ1 can go down to 46MBs with these flies and my drive is right there. The problem I've found is where worst case with large sequential should be at 121 MBs I get 80-81MBs. the data stream first spikes at 121, where it should be, but immediately drops to about 80 MBs or 2/3 the speed or even lower (see speculative). It appears the device cannot handle the compression and or other things it needs to do and collapses.

I should note the first few hours of my owning the drive I had 140 MBs on the write speeds of the highly randomized data. At first I thought just a system restore killed my speeds but I have diagnosed it further.

Others showing this issue have just been brushed off so far, I will not be. I have diagnosed my system and the software bus and trim are of a non-issue etc.

Now admitedly I am feeding the SSD from a HDD capable of feeding the drive at 100+ MBs. Those just feeding the drive by DVD, network or USB 2.0 etc may never notice this. If you plan to use though SATA, ESATA, USB 3 or other high bandwidth interface this could be an issue.

WARNING; you may not already have an issue to begin with and doing this could then cause it by creating the high speed randomized data..........
If you use CMD to benchmark the drive and get lower than the "worst case scenario" as in the link above you may have issues as well.

Now this may be just a one off issue but I have seen a couple of other posts with similar issues. So this is to get the word out. It took me so far three nights of fighting with this.

Speculative;

I highly suspect there is a down clocking provision of the compression engine going on here. My 80 MBs is 2/3 of the original speed and I do see the transfers start off at 121 MBs. Also I will see, while monitoring the transfers, the speed jump in steps of 60,40 and 20 MBs and sometimes back up to 121 MBs.

This could mean variable performance and worst. It could mean over time the perfomance degradation goes even further............

If this is true the other end of the problem could then be since it is a designed provision we may have no recourse when having a drive with worstening performance issues.

I appologize for the long RANT............

 

I dont think collapse is the right word, the thread title lead me to believe I was going to read about a drive failure.

What you explained though is not a failure but just a performance degradation.

As with any new technology if your one of the first on the buss you have no idea where the ride is going to take you. So thanks for giving the heads up about the performance issue.

 

Yeah the title should be performance collapse, Thught I had editied it to that now I see I didn't. Up way too late and way too tired.............

 

I finally got my answer back from Mushkin on the SandForce 1200 and the Callisto DX’s. Apparently there is a load sensor within the hardware. When it senses the compression loads are high for a comparatively long time it slows the NAND to preserve it. The recourse left is to either wait an impossibly a long time or to do a secure erase and start over.

I can say the secure erase method does work to restore the drives write performance. After 60 plus hours of logged off time it did not restore the write performance at all. I have asked if we can get a firmware with a more aggressive regeneration capability and/or one that allows for less aggression in slowing the drives NAND down.

My feeling is in a consumer device highly compressed files of picture, video and music etc. are common place. So while this slow down in an unattended enterprise SF-1500 drive on a server has its place it doesn’t so much in a consumer SF-1200 device, at least with its present implementation.

So in all at least it is not a heat related issue it does seem to be load oriented issue as I had originally forecast. Hopefully it can be addressed in the future as well. For now I will be rethinking my workflow to try and keep those files away from the drive and whatever you do, do not benchmark your drives with CMD (Default Settings), AS SSD or similar as these will for now prematurely stress the compression and subsequently hasten the time for the slowdown of your SSD’s writes.

The only safe benchmarks for these would be either ATTO or CMD only using either 00 or FF and not the default “Random” data…………….