Jump to content
RESET Forums (homeservershow.com)

RAID. TRIM. SSD.


dvn
 Share

Recommended Posts

OK. I'm not seeing any speed improvements.

 

For what it's worth, I also tested the idea that a free space wipe would trigger TRIM. If only... That also seems to be a 'phail'.

 

*Remember, I'm testing these things in RAID0 to verify TRIM support promised by a beta RST driver. I have no idea if TRIM will actually ever work in my RAID0. Maybe I should try these tests with a single drive config first, right? Or maybe I should SE the SSD's, restore the image, and just run my system normally for a period of time (how long?) to see if write speeds degrade. That may actually make more sense. Anyway...

 

It occurs to me that I'm not really sure of the extent of TRIM capabilities. I mean, I know TRIM is triggered by file deletion, but does it run when I do something like defrag an SSD drive? Will it be triggered by a mass deletion from a free space wipe? We all know the cautions against running defrag on an SSD. But is this a caution because it degrades write speeds until TRIM deals with the mess? Or is this something outside the scope of TRIM's function? From my admittedly unscientific testing, I'd say that defrag's actions do not trigger TRIM, which is probably one of the reasons people advise against it.

 

Also, I gather that TRIM acts virtually instantaneously in the case of a singe file deletion. Is there a limit to the size or number of files that TRIM can address? Probably not, but if not, why doesn't it appear to provide a speed boost when you run a free space wipe with the intent of restoring degraded drive write speeds?

Link to comment
Share on other sites

  • Replies 60
  • Created
  • Last Reply

Top Posters In This Topic

  • dvn

    28

  • ImTheTypeOfGuy

    6

  • pcdoc

    12

I decided to run the RAID0 for awhile, give this beta driver a chance to show me something. And I'll hold off on running anymore benchmarks until next weekend because too much of that can degrade performance. Geez, lots of picky little things with SSD's, eh? It's like 'stand on one leg, touch your nose, recite the alphabet backwards...' haha

Link to comment
Share on other sites

Agreed. That is a big reason why I'm still skeptical of SSDs. I do have to say that I recently checked Intel's site, and I found an updated copy of their SSD Toolbox. It seems like they recently added a checklist app that looks at Windows for optimal settings. It turns out I must have missed something because it found Superfetch to be enabled on my machine and even offered to fix it for me automatically. I'm not seeing much of that from other manufacurers...

Link to comment
Share on other sites

Intel does have the best suite of applications and support however, with SSD's we end up messing around too much as it is a new technology. The fact is we are enthusiast and want to get the max out of our systems, but in the real world, you really only need to install it and run it. Most SSD based store bought systems do not have these tweaks and run just fine. I would not be skeptical of using an SSD as after you put one in you can never go back, what I would do is keep in perspective. It you got them for tweaking purposes that is one thing but if you got it to just run, you will be perfectly happy with it as it will run circles around you mechanical drive with very minimal efforts.

Link to comment
Share on other sites

Exactly. And if the performance drops off significantly, an image>Secure Erase>re-image will put the drives back in top condition. As to SSD tweaks, I downloaded and ran SSD Tweaker. I believe this takes care of all the tweaks that have been mentioned on BYOB. Since I have 8 GB RAM, I've done away with the page file and I've disabled hibernate. Between the two, I must have recovered over 10 GB of wasted space.

 

@mrossco - How are you liking your Corsair SSD?

Link to comment
Share on other sites

The difference between perceivable performance increase using tweaks vs. running as-is is just too big to ignore. It looks like HP is going to bundle their Slate 500 with an SSD, and I know others are starting to do the same. If I could make my device seem to run that much faster with little added cost, wouldn't I want to find a way to bundle OS config changes with my device?

 

@mrossco - How are you liking your Corsair SSD?

 

So far, I like it. I use it on my workstation at home, which runs Windows Server 2008 R2 so that I can leverage Hyper-V to run work-related 64-bit VMs. The Corsair speeds up my system enough to tied me over while I save $$$ for my next build.

Link to comment
Share on other sites

...I recently checked Intel's site, and I found an updated copy of their SSD Toolbox. It seems like they recently added a checklist app that looks at Windows for optimal settings.

Could you post a screen shot of that app? I'd like to look over that checklist to see if it includes all the things that SSD Tweaker does.

Link to comment
Share on other sites

Well, live and learn. Benchmarking apps such as CDM which write around 10GB highly compressed data per run have a highly detrimental impact on SSD write performance when run repeatedly over a short period of time. People who have discussed this say that if one must run CDM, run it using the 100MB test size and limit the number of test runs per test to 3. Run no more than 3 tests at a time and no more than 1 test session every couple weeks. This is info was gleaned from OCZ forums. If what some folks have said about the way nand drives operate is true, SSD's basically use every available data block on the drive before writing again to the same blocks. So running CDM 4 or 5 times on a 60 GB drive...you can see where this is going.

 

The good news is that Secure Erase restores a drive to its initial pristine state. The warning, of course, is that this is destructive to all data on the drive. But that's a given.

 

So I'm now rethinking how I've gone about testing the beta Intel RST driver to evaluate for presence of TRIM. I think I should probably do the image-SE-reimage loop once more and just use the drive as I normally would. Restrict CDM benches to once a fortnight.

 

ATTO is a different story. It turns out that ATTO is the preferred benchmark tool for manufacturers because it uses highly compressible data which Sandforce controllers handle extremely well with little to no affect on write performance. That basically means that ATTO is the best bench to use if you want to show off your drive's performance.

 

As to whether TRIM is actually working with this beta RST driver, I cannot say for certain but I believe it just might be. Restarting the evaluation process as I mentioned above should give me a more definitive answer, but this will take some time. How long, I'm not sure. Since I'm inexperienced when it comes to what to expect for performance falloff with normal usage, I'm not really sure how long of a period I'd need to keep monitoring. If anyone has an idea as to how long it takes before one might expect to observe the effects of TRIM, let me know. I suppose in a perfect world, I'd have 2 identical computers/setups, running identical scripted tasks over a period of weeks, with one computer in IDE mode and the other in AHCI, periodically testing the performance.

 

Tweaks mentioned by pcdoc and others are things to consider. I'm particularly interested in the relocation of Firefox cache to an HDD. System tmp and temp folders whose paths are defined in Environment Variables will also be moved to an HDD.

 

As to Dave's question of whether he should grab another 60 GB and set up a RAID0, I'd say sure, why not. Have a regular backup scheme (which you already do) and you should be good to go. If things don't work out, you can always run SE and redeploy your SSDs in another setup.

Edited by dvn
Link to comment
Share on other sites

Guest no-control

Sorry guys I gotta step in as I have a different experience and opinion.

 

First off let's define what TRIM is

 

To Quote Wikipedia

TRIM enables the SSD to handle garbage collection overhead, that would otherwise significantly slow down future write operations to the involved blocks, in advance.

 

Now with that said EVERY SSD has a garbage collection Feature that goes in and cleans the cells to optimize write speeds.

 

TRIM not being supported in RAID is totally true, HOWEVER, the TRIM command isn't necessary if you're willing to sacrifice around 10-20% of total storage capacity to allow the drives' built in Garbage Collection routine to do its work automatically in the backround with RAID. Anandtech tested this a while back, works just fine. You just have to leave a portion of each SSD in the array unformatted and unused for it to work. As i said above all SSD's have this feature in the controller's firmware. BTW this feature is why the Sandforce drives are ~10% smaller than previous generations of drives (my Falcons are 128gb, but the Phoenix pros are 120gb Indilinx vs Sandforce ctrls) its to provide room for the GC feature to work.

 

This is from Tony, the support staff on the ocz forums;

"Unplugging the drive but keep it powered allows the drive to enter into its recovery period...if you bench the drive to death for 6 hrs and write 6 days worth of writes in those 6 hrs the drives with 0 writes per day there after will probably take anything upto 6 days to recover to full speed. If you continue to use the drive applying light load the drive may take 10 to 12 days to recover."

 

From the research I've done and conversations with OCZ & G.Skill support staff say that their GC works even faster if it isn't connected to the computer, but it still works no matter what. He also gives a very high, if not pathological work load, as an example (ie. "six days worth of writes in 6 hours") of performance. If OCZ's avg. expected daily work load numbers are anything like Intel's (~5GB of writes a day averaged out over several years) then I wouldn't worry about RAID with their SSD's either.

 

If anything GC theoretically should work better in RAID since you would be splitting the write load between multiple disks. Anand demonstrates it pretty well in the benches he did (see link above). Note that Anand used Intel SSD's and he didn't have to do anything silly like leave them unplugged from the computer for days on end and such. So long as he wasn't doing anything that did heavy random writes while using the computer the drives' built in GC was able to seamlessly work in the background to maintain performance. In order to even get any performance degradation in the first place he to do pathological work loads, and GC still was able to fix things just fine quickly once he stopped flogging the disks.

 

On an SSD write performance WILL degrade over time, and the amount of performance degradation is proportional to the amount of random small writes that are done during the use of the SSD and the amount of data stored on the SSD. But since it's NAND flash running in several parallel channels that performance degradation doesn't get noticable for a loooong time. Measurable with benchmarks? yes..... noticeable in everyday use? Nope

 

One simple and rather obvious fact about the whole TRIM issue is that it ONLY affects write speeds. The vast majority of people who use an SSD are going to focus on improved boot time, improved program loading times, etc. That all involves reads from the drive almost exclusively. Therefore no direct correlation to TRIM's primary goal of increasing write speeds.

 

My opinion is to not even worry about TRIM. Use the SSD as you would any other Hard drive. Taking the time setting up your software to minimize writes if at all possible is great due diligence but not required. This is why having a secondary spindle data drive is a good idea. Odds are you will be able to use the SSD for years without a noticable performance degradation regardless.

 

I've run my SSDs in a RAID 0 arrays for well over a year and a half now and notice no lag or slow down without TRIM.

 

Again just my experience and opinion.

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
 Share


×
×
  • Create New...