Announcement

Collapse
No announcement yet.

Why You Need TRIM For Your SSD

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Why You Need TRIM For Your SSD

    here is an excellent article concerning TRIM and its impact on SSD performance...the article also discusses the different firmware revisions and controllers...performance with Indilinx based drives (like the G.Skill Falcon) are improved greatly with TRIM enabled firmware (FW 1916)

    http://www.bit-tech.net/hardware/201...nce-and-trim/1

    now if only G.Skill would release firmware 1916!!

  • #2
    Thanks for posting that. I gave it a quick scan and it does look good. I will give it a more detailed perusal later.

    While Trim is certainly good there are a couple of points that need to be made. First garbage collection without trim will achieve the same ends. Garbage collection and trim is just more efficient at it. Secondly in a Raid setup due to the write caching the performance gains using trim will usually not be noticeable in practice. Check out:
    http://communities.intel.com/message...ode5COMS#24652
    'I actually do understand the issues of TRIM and SSDs fully. I've posted several times about this topic on past threads. However, I'll emphasize my point, too: The gains you get from using TRIM are minimal in real-world use. Sure, you can see slight differences by running benchmarks, but odds are the system will not operate this way in real-world conditions. RAID benefits significantly by having a built-in cache, thus the difference between enterprise-level controllers and 'low-cost' home consumer controllers. Gaining RAID performance isn't just a matter of 'going wrong' but a matter of using the right combination of hardware. It's a proven science, not open to much theory. A RAID cache will benefit your write performance more than the TRIM will, and that's a fact of I/O, period. If you wish to wait for TRIM to be fully implemented, that's your choice, of course, but I will guarantee your wait will be rewarded with very little satisfaction, except perhaps on the cheapest of SSD components. However, I only speak for the majority of systems operating out there. Perhaps you have a very unique situation where you're constantly writing gigabytes of data on an hourly basis to your SSDs. But, if it happens to be for your system pagefile or temporary files, perhaps an additional RAMDISK would be the next best upgrade to your system.'

    While I certainly hope they release 1916 forthwith I wont be flashing to it straight away. That day will probably come when I do the next major upgrade to my machine which is likely to occur when fusion i/o gets their fusion extreme product bootabe and I relegate my raided falcons to storage rather than boot duties.

    Thanks
    Bill

    Comment


    • #3
      Again thanks for posting the article. To be frank I don't quite know what to make of it. I have read extensively on SSD performance degradation over time and every single article I read stated reads were virtually unaffected eg:
      http://it.anandtech.com/showdoc.aspx?i=3531&p=14

      Writes were the thing that was supposed to take the big hit. And that can, in practice, largely be overcome by write caching. Because of that I was not that perturbed by the issue.

      However that article shows large performance hits to reads with time. That will not be masked by caching. Exactly what causes it I have zero idea. Yes the drive becomes heavily fragmented with time so read will slow down a bit and that was what other articles showed. But the slowdown was negligible.

      All that I can think of is that it was a byproduct of the particular way they prepared dirty drives. I did read an article of someone who deliberately devised a way to really foul up SSD's with massive hits to both reads and writes. But the article was at pains to point out that it was deliberately contrived and is very unlikely to occur in practice.

      From my point of view I do run both ATTO and AS SSD every month or so. So far I have seen zero degradation in my raid set up. Previously I was going to wait until my next major upgrade to flash to the latest firmware but considering the article I now think it is much more reasonable to keep an eye on its performance and flash when reads degrade. Hopefully it is a byproduct of how they prepared dirty drives.

      Thanks
      Bill

      Comment


      • #4
        Regarding the fact that it seems like G.SKILL is late on releasing 1916:
        Originally posted by GSKILL TECH View Post
        We continue to test and improve our firmware. Unfortunately we do not participate in the firmware release competition, but we are doing our best to provide users with a stable, bug free firmware, as soon as possible. We are running thorough tests to fix bugs, rather than having users do the testing for us. We appreciate your patience and continuous support. As always, we will post new information and updates here for everybody.

        Thank you
        GSKILL SUPPORT
        This was taken from the thread When the firm 1916???.

        Comment


        • #5
          Hi All

          In light of the fact the posted article on performance degradation did not seem to gel with what others such as Anandtech had previously detailed (namely reads were largely unaffected) I decided to research it a bit further and now believe I understand what going on.

          If reads are affected depends on your pattern of wear. If you filled up your drive with tons of small random 4k reads and writes then very little contiguous space will be left. It is this contiguous space that is required to have the really fast large file reads. If when you read a 4k block you can rely on successive blocks containing the data then you can get them at the same time and speed up reads - it will be in the buffer when required. If that is not the case then a lot of random i/o needs to be done to read a sequential file which will slow reads down considerably.

          The way Annantech and others created dirty drives was writing a large contiguous block to it then delete it. This will leave large read speeds unaffected. The way the posted article prepare dirty dives did not do that but instead created a highly fragmented drive with little of no contiguous space. Storing and then reading a large file on such a drive will be slowed considerably. However it should have zero effect (and the posted article bears this out) on the highly critical random 4k read.

          Bottom line here is sequential reads of large files will be impacted but what really matters for SSD performance in loading programs etc is the random 4K read speed - which is not affected. The key performance indicator here is the 4K-64 thread read speed supplied by AS SSD. Since I use raid mine is a whopping 120mbs - even better than a single Intel (but not by much - since its so critical Intel optimizes that at the expense of sequential reads - which is why Intel tents to feel perkier than other drives). That why my drive feels perky and why I suspect it will continue to be so even when well used. I simply don't muck around much with large files.

          Of course I will continue to monitor its performance and if it degrades significantly flash to a new version.

          Thanks
          Bill
          Last edited by bhobba; 02-04-2010, 08:01 PM.

          Comment

          Working...
          X