Long-term performance analysis of Intel Mainstream SSDs

[flynace]

"There Is No Defrag (yet)...there is currently no ‘easy’ way to defragment the sub-block level fragmentation that occurs as a result of write combining."

Isn't this the problem that Diskeeper's HyperFast module for SSDs supposed to solve?

During your (PCPerspective's) testing, can you see if using HyperFast recovers the SSD performance?

It would be interesting to know if a solution is out there, albeit at an additional cost, until Intel provides one.

[malventano]

The hyperfast page over at diskeeper makes it sound like its the best thing for SSDs, but unless it can do so with relatively large writes, it will only slow down the X25-M (like every other defrag did). Our tests of the Intel drive showed the free space available is irrelevant, especially after a total of 80GB has been written to the drive (80 GB worth of writes, not filling it to 80 GB capacity).

They guys over at the OCZ forum gave it a shot as well and saw negligible improvements. There is also some gouge there from diskeeper, saying to only run it an hour or two at a time. It does not appear diskeeper is as "SSD aware" as their hyperfast page would like to suggest.

[jmke]

Do you guys know where to get the HDD Erase version mentioned in the review which does work with Intel's SSD ?

[malventano]

We are hosting the older version and have linked it off of the article page talking about it.

[jmke]

I see you've added it, thanks!

[ForumUser]

There is another piece of software that may be of interest in regards to this issue: http://www.easyco.com/home/index.htm

Much like HyperFast, it claims to improve SSD performance and lifetime by ludicrous amounts. It does sound pretty similar, but since DisKeeper doesn't explain how their product actually works, I can't really say.

I've so far avoided trying out the free trial of EasyCo's MFT because it results in a major modification to your system (it creates new partitions and clones around your OS when you install it), so I imagine it's only worth the hassle if you plan to buy and keep using it for a long while.

Then, would this product be a useful cunteragent to the fragmentation caused by write combining (I'm guessing the answer is no, but I'm no expert)? Even if not, though, would you generally recommend the use of software like that, in combination with for example Mtron SSDs? It might be a good subject for a whole new article, as well.

[mmaterie]

HyperFast is a great solution for most SSDs. The Intel X25 is the only SSD to date, that we know from testing does not benefit from NTFS free space consolidation - for the reasons you covered early in your article. We work closely with Intel and know these drives well.

It's important to note that you won't see dramatic differences optimizing an out-of-the-box SSD with HyperFast, for the same reason you wouldn't see huge differences defragmenting a relatively unused HDD. The key to HyperFast is fixing the fragmentation of free space (in the file system not in the SSD itself -btw) that naturally occurs over time, especially when an SSD is used as the system partition. In the case of the metrics we provide on the website and in papers, the test case was a Vista system after about 6 months of normal use. The results published in the OCZ forum did note improvements, and their results would be proportionate with the free space fragmentation levels in those tests. If they only started with a small amount of free space fragmentation they'd only get a small performance gain.

PS: The gouge was a recommendation, re-published (by a poster), that was initially provided by a Tech Support rep prior to the release of the new SSD-aware products (which includes HyperFast), and well prior to the X25.

-Michael Materie
Diskeeper Corporation

Originally Posted by malventano

The hyperfast page over at diskeeper makes it sound like its the best thing for SSDs, but unless it can do so with relatively large writes, it will only slow down the X25-M (like every other defrag did). Our tests of the Intel drive showed the free space available is irrelevant, especially after a total of 80GB has been written to the drive (80 GB worth of writes, not filling it to 80 GB capacity).

They guys over at the OCZ forum gave it a shot as well and saw negligible improvements. There is also some gouge there from diskeeper, saying to only run it an hour or two at a time. It does not appear diskeeper is as "SSD aware" as their hyperfast page would like to suggest.

[malventano]

Originally Posted by mmaterie

(excellent info)

-Michael Materie
Diskeeper Corporation

Thanks for the info Michael. I suspect the guys over at the OCZ forum were testing things on a fairly 'fresh' drive with little benefit to free space consolidation. It does make sense that consolidating free space will help, especially write speed, as NTFS likes taking a big file and scattering it across every small gap along the way, which would make a (regular) SSD crawl.

Please contact me via email to discuss further.

Thanks again for taking the time to reply here.

[hp79]

Are there any updates or related articles planned about the X25-m? PC Perspective is the only place that wrote about this fragmenting problems with intel x25-m. I'd like to get any updates if there are any in the future.

[malventano]

Originally Posted by hp79

Are there any updates or related articles planned about the X25-m? PC Perspective is the only place that wrote about this fragmenting problems with intel x25-m. I'd like to get any updates if there are any in the future.

We did all we really could for the moment. The article gets the data out there and hopefully curbs peoples' usage to avoid the worst case.

Here is a bit I posted on another forum in response to a question of the X25-E showing the same fragmentation:

We briefly had an E series drive to test, and it does fragment (by about the same ratio seen on the M drive), but it immediately snaps back, just as the M series drive does most of the time. We only had the drive for a week so we were not able to find out if it hits the 'point of no return' we saw with the X25-M.

Keep in mind this fragmentation is really just a down side to write combining. It is a simple tradeoff. The X25 drives achieve crazy high small write IOPS using this technique, in fact the M even beats out other SLC drives in this very area. If you do nothing but hit it with small writes, it is eventually going to slow to the point where it acts more like a non-write-combining drive, and larger writes will take a hit because of the fragmentation that came about from the initial write combining. Hitting it with a mix puts the average write speed at some point between 40 and 80 MB/sec, all depending on the particular mix of large and small files written (i.e. your usage pattern).

Even after writing the article, I still use my X25-M as my primary OS drive, even with all of the other SSD's at my disposal (including SLC units). Its insanely high IOPS performance makes it the best choice. You just have to realize that hitting it with a constant power user workload will slow it down. It is *supposed* to bounce back with larger writes, and we are working with Intel so they can easily replicate the 'bad' cases we saw. I suspect that once the Intel techs can dissect what we stumbled on, that a simple firmware fix will solve that particular issue.

A side note: Some sites have taken the Intel statement and spun it as if the drive does not slow down at all. This is not accurate, and I believe does everyone a disservice. Intel and PCPer have a mutual understanding that usage will drop performance to a steady state value based on that usage. You have only to read the "Ask and Intel SSD Engineer" article on [H] Enthusiast to figure that much out... It is also worth mentioning that the official write speed spec of the X25-M is 70 MB/sec, where a 'new' drive will write at 80/sec (closer to 90/sec with AHCI enabled). Clearly Intel considered this slow down when they published the spec, and it was very honest of them to do so.

I'm holding off on another X25-M article for the time being, mainly because we are working with a few groups behind the scenes:

- Intel. We provided them some usage pattern data to assist their techs in replicating the more 'aggressive' patterns we threw at our drive. We were able to achieve this condition with 'real use', and we also provided them with a synthetic method we found to cause severe fragmentation in less than an hour.
- HDDErase. It turns out a simple bug in 4.0 was preventing its use with the X25-M. We are working with the CMRR guys on that as well. Their funding was pulled some time ago, but they might still come through with one last bugfix build.

With the possibility of an X25-M firmware update / HDDErase update, it seems worthwhile to reserve our efforts until there is more to report on the topic.

[heavyharmonies]

*sigh*

My X25-M that I'm using as my boot drive in my desktop started halting today, and sure enough it's doing the dance of death.

Ugh.

I wish there was a simple fix that didn't involve zapping the drive.

[malventano]

Originally Posted by heavyharmonies

*sigh*

My X25-M that I'm using as my boot drive in my desktop started halting today, and sure enough it's doing the dance of death.

Those results are curious - Your write speeds look pretty good. Those particular results suggest something other than fragmentation may be causing your issue. It seems to only hiccup in the higher read speeds, and when it does, it *really* does, suggesting a timeout issue.

Try swapping to a different SATA cable and if that does not help, verify nothing has changed as far as your storage drivers go.

[BWM]

Please try setting up the ATTO options as shown in this pic then re-run the tests. The new results should be quite different.

[Xetick]

Interesting article. It makes me question if I really should get a X25-M 80Gb drive.The plan was to get 2 X25-M 80Gb drives and put them in a raid 0 config to get the read and specially the write speed up. I guess in theory it should also double the life time of the drives since only half the data is written to them.

However what I was wondering is if these tests are done with the SSDSA2MH080G1 or the SSDSA2MH080G1C5 models of the X25-M 80Gb. I have search the net to find out the difference between these two models without any success so could it be that they have addressed this exact issue with a new drive release? The only difference I have found is that one is a bit more expensive then the other.

[graye]

I've read the entire article, and I still can't figure out how the Intel SSD *read* performance drops when the wear algothrym gets messed up beyond repair. I can understand the write performance dropping... but not read.

BTW: Somebody could make some money if that'd create a piece of software that will "reset" the DCO area of the drive without distroying the data contents.