Solid state drives were first released to the consumer in the latter part of 2007 to which I was fotunate to be one of the first to publish a SSD review that is still available on our site today.
At one point in the past few years, it became common knowledge in the industry that solid state drives would suffer performance degradation with age.
I found this very unusual because, for the most part, I had not experienced this with any of my drives and believe me, they have all been around the block a few times in the past few years. I have tested drives from pretty much each and every manufacturer including form factors of mSATA, SATA, mPCIe and PCIe.
PERFORMANCE DEGRADATION CONSIDERED
In an effort to find out why my drives were running so well, I examined what I was doing different than the average consumer and, more specifically, how my system setup could contribute to this. Three things stood out. I have been running 64 Bit on my system since before it was even believed to be viable. Would you believe I had a small part in the push for Dell to introduce it to the market and also, people used to laugh and think I was ridiculous to believe it would ever make it?
The other two variables were that I don’t run a Pagefile and haven’t run System Restore since my first SSD in 2007. Simply, I believe Pagefile was created for systems that didn’t have adequate RAM and we have allowed it to become a crutch now that RAM is affordable. I also I found it much easier to simply backup regularly than allow System Restore to do its magic.
In considering all points, I was a bit confused except one thing stuck with me. System Restore grows as it backs up your system and changes your storage characteristics as it continually establishes new restore points. This seemed like a common sense approach to think it might be related to System Restore..
THE BIG DISCOVERY
I have been watching SSDs for some time and observed that non-TRIM SSDs such as the Intel X25-m G1 were particularly prone to this phenomenon of degradation which continued to be present through the introduction of the Intel G2 with TRIM. I purchased and installed a Intel X25-m G2 160Gb SSD. In the process, I elected to leave System Restore on because, well, I was about to rip apart the OS in order to see how it ran best with a SSD and would want to be jumping back and forth.
I am also a performance nut and rely heavily on Crystal Disk Mark as it provides one with both large sequential as well as 4kb random performance evaluations for your SSD. Two weeks after playing around with my system, I observed a large drop in the performance of my SSD. I wish I had kept the original result so I could post it here but no fear, I will have a test for you to try later on. Here is what Crystal Disk Mark looks like:.
Parallel to this, I had also installed and used the Intel SSD Toolbox Optimizer twice and observed both times that it took an extremely long time to manually do what TRIM was supposed to be doing. Quite frankly, it took over an hour. I wondered, in fact, if TRIM was working at all.
CAUSE AND SOLUTION
I subsequently tried everything to no avail and, only after I shut off System Restore and the file was deleted, did I see a sudden jump in the performance again. Performance returned to where it should be in fact. My next trick was to try the Intel SSD Toolbox Optimizer. I did and it took a split second as I was always informed that it should. Maybe we have hit on something here!?!?
Is it possible that the way System Restore allocates its restore points makes it difficult or even prevents TRIM from running properly?
Its been several months now and my system is purring like a kitten. I have absolutely no slowing or performance loss of any kind and the Intel Optimizer continually works in a split second to finish its job. I subsequently put the thought out with Intel where it went unanswered and have posted with respect to it on approximately 10 technology sites. I have to concede that I have yet to have anyone confirm this scientifically other than most, including Intel, jumping on the train and recommending that System Restore be turned off with a SSD. Go Figure!!!
What I have garnished is a million thank you’s along the way from SSD enthusiasts who discovered that this was the cause to sudden slowing observed with their SSDs. In fact, I have yet to have the theory disputed and I felt that Intel might be the first to jump on this.
Here we go. It is simple.
- Download Crystal Disk Mark and do a test first and foremost;
- If you have an Intel, download the Intel SSD Toolbox and optimize. If it take a long time you now have an idea why. For non-Intel drives you can do much the same with any optimizing utilities are available for your SSD;
- Turn off System Restore. It has to be completely shut off in order for the files to be deleted from the SSD. It cannot just be reduced;
- Optimize again. If you have the Intel and it is very quick now, you are on the verge of a big discovery;
- Take another Crystal Disk Mark Score. If there is a big difference, you are not alone because many people have told me that this was EXACTLY the case; and
- If there is no difference, don’t be afraid to turn System Restore back on if you need it.
Alot of confirmation has occurred to enable my writing of this article yet there is very little education or confirmation about the web. I would believe it is simply a matter of Intel, or any other SSD manufacturer, not wanting to ask one to shut down a system resource that has always been considered so valuable.
If you elect to follow the test procedure, please comment one way or another. I am very interested with any further proof or constructive thoughts that may prove or disprove my theory. As it stands, I have been waiting a very long time for an answer from Intel, yet many have thanked me for helping them discover why their SSD has slowed since installation of the OS.
After switching from IDE to AHCI with my Intel X25-M 80GB, I got 2x worse random 4K (r ~ 10MB/s & w ~ 12MB/s – compare to yours :)) with QD=1, and got a boost for QD=32, ~5x for reads and 2-3x for writes.
Still, using HDTune 4.5, I currently have <4000 4K random read iops, which is disappointing and not even close to specs.
Tried your test, didn't work – same stuff.
I didn’t see much of a difference with system restore on or off. I did the test and I should add that I did disable the system restore option before turning it back to to do the test.
Test with sys restore on:
With sys restore off:
I just tried this on my AMD Phenom II x6 system with my new Intel 320 series SSD, 128 GB. Though the Intel SSD Toolbox optimization was significantly faster the second time (after System Restore was turned off), the CrystalMarks taken before and after showed no great improvement. The biggest improvement was a 10% increase in 512K Write, but other than that the differences were negligible (+-0.5%).
Unfortunately this theory has not been proven with my setup. I have a non-intel SSD, Crucial M4, and my before and after Cyrstal disk mark results are within 2% when disabling restore point.
I went even further and tested to Restore on for 1 week, ran crystal disk mark and recorded the results, disabled System restore and after a week ran crystal disk mark again and the results show no definate improvement with this “tweak”.
Only downside i see is the extra space allocated by System Restore, but the benefits of rolling back from a shotty driver or app weighs more IMO.
USing windows 7 x64 so i take it the TRIM feature works well!
SITE RESPONSE: Thanks for the comments and this article is somewhat dated, albeit valid as shutting down System Restore is a thought held by many, including some manufacturers.
i’ve also had a small increase in disk performance (in some tests even a small decrease) when i dissabled restore points.
using ocz agility 3 – 120GB and running windows 7 x64. i will turn it back on and redo the tests in a week or so and later as well and if i see decrease in performance i will repost here and let you know that in fact it does matter much (since the system is quite new and maybe that’s the reason that decrease is not yet well visible)
same or worse results here
intel i7, 8gb ram, intel 510 250GB SSD, Win7 pro x64
Ran 1st test while i had word/excel/outlook/7 tab ie session all open, and ran CMark software
ran “AFTER” test with fresh boot from power off – results overall were – WORSE not better …
only exception was that intel optimizer did run in ony 3 seconds, instead of 10 seconds in “BEFORE” test …..
Running Win 7 x64 with Patriot Wildfire as main drive and Patriot Inferno as secondary SSD. Minimal difference in the tests, so I have turned System Restore back on
While you indicate that you do not use a Pagefile, how does one accomplish this? I just installed my first SSD in a laptop as a “dry run” and will be installing another (Samsung 840 Pro) in my main system. Can you tell us how to eliminate the Pagefile?
Go to our SSD Installation Guide and it will help you. Havent used pagefile in several years now…not a problem yet…
I have not used system restore for a long time as it often didn’t work when I needed it. However I did try getting rid of pagefile as I have 8 GB of RAM and very seldom do any really intensive stuff. I found that my Boot time went from 15 seconds to about 28 seconds and it also took a little longer to do a system backup with Macrium Reflect. As soon as I allowed a Pagefile again the previous performance was restored.
I wonder if anyone else has noticed this.
How about turning off/deleting system restore, optimize, and turning back on again. It has saved me a few times.
Interesting, There is a program I found and use called RESTORE POINT CREATOR which I utilise to make staged restore points covering ,say, ten Windows restore points. I find this very useful in not having to reinstal win upgrades if I have to jump back to a much earlier restore point. I wonder if it will still work if the Windows restore system is disabled ??? —- If it did then one could have the best of both, SSD performance, as well as Restore Points. —- I am passing this to you because you have wide experience with this subject and I still only have HDDs. Please reply with any result you discover, as I said at the start— Interesting !