Sign in with
Sign up | Sign in
Your question
Solved

Three questions on SSD (partitioning, SATA2 and writes)

Last response: in Storage
Share
April 11, 2011 8:10:11 PM


I am likely to buy at least one Vertex 3 SSD drive, but I have some questions:

1. At SSDs today, is it still a problem with writing to heavy to the drive? Is that something that should be avoided? If so, how big is the problem?

2. I have got some suggestions to leave a certain part of the disk unpartitioned, perhaps about 20%. That might be reasonable, but I would like to hear if that is a general opinion that such a partitioning is wise?

3. I might have to use a SATA2 connector to the SSD if using it as boot device. Since the Vertex 3 promises speed at 500 MB/s this would not be possible at SATA2. However, since such numbers (which are extremly cool) does not represent typical usage, perhaps not even heavy usage - is it likely that it could be a "limit"? I assume that what matter most is the read and write performance of small (4KB/8KB) random IOs and would SATA2 in some way be a worse connector than SATA3 for that?


I am grateful for answers, comments or suggestions for any of these questions.
a c 415 G Storage
April 11, 2011 10:53:55 PM

1) There are two issues with writes - they are typically slower than reads and they "use up" the life expectancy of the drive. SSDs have a limited number of writes they can handle. I don't know about the Vertex 3's, but the 160GB Intel X-25M G2 drive is supposed to be able to handle around 5GB/day worth of write for 5 years or more. So the answer is generally: try to avoid putting files with heavy write activity on your SSD, but don't stay awake nights worrying about it.

2) My personal preference is to partition the whole drive but try to keep at least a certain percentage unused. That's similar to leaving some space unpartitioned, except that in a pinch you can use the space for a while if you really need to. The reason you do this is to leave spare memory available for the SSD's controller to do "wear leveling" in order to balance the writes across all of the memory cells and extend the life of the drive.

3) You'll loose some peak transfer rate performance, but the biggest benefit of an SSD is it's very fast access times, and those won't be hurt by SATA 2.
m
0
l
April 11, 2011 10:55:38 PM

ricno said:
1. At SSDs today, is it still a problem with writing to heavy to the drive? Is that something that should be avoided? If so, how big is the problem?


Please clarify, what do you mean?


ricno said:
2. I have got some suggestions to leave a certain part of the disk unpartitioned, perhaps about 20%. That might be reasonable, but I would like to hear if that is a general opinion that such a partitioning is wise?


I can't think of a good reason to partition a drive, not really necessary.
m
0
l
Related resources
April 12, 2011 5:09:15 AM

sminlal said:
SSDs have a limited number of writes they can handle. I don't know about the Vertex 3's, but the 160GB Intel X-25M G2 drive is supposed to be able to handle around 5GB/day worth of write for 5 years or more.


Is there any term for this kind of "write maximums", that is - can I look at the specification for a SSD and find how many writes it can take (as the around 5GB/day in 5 year)?

I guess also it is kind of hard to guess how many writes a typical system does, but that could be studied by Performance Monitor. Could be interesting to see a typical day, but I suspect quite low.

As to this write wearing, will it end up to the SSD being damaged/not working if doing way to much writing (say 100 GB write/day very often) or will the performance decrease?

sminlal said:
2) My personal preference is to partition the whole drive but try to keep at least a certain percentage unused. That's similar to leaving some space unpartitioned, except that in a pinch you can use the space for a while if you really need to. The reason you do this is to leave spare memory available for the SSD's controller to do "wear leveling" in order to balance the writes across all of the memory cells and extend the life of the drive.


So there should be no technical difference between unpartitioned space and empty logical space inside a partition? What about erased files (removed from the recycle bin), on a traditional drive they remain on the disk sectors until overwritten?

sminlal said:
3) You'll loose some peak transfer rate performance, but the biggest benefit of an SSD is it's very fast access times, and those won't be hurt by SATA 2.


So there is no other differences between SATA2 and SATA3 than the throughput itself? Then I can live with the SATA2 connector for the boot drive until I (at some time) upgrade to a new motherboard with only SATA3.

Even as I do think it is very very impressive with these numbers of fantastic sequential transfers, it should be that often the 300 MB/s limit should be reached.


m
0
l
April 12, 2011 7:59:11 AM

There is about 5 mlns write operations allowed for a SSD before it will go to the trash. So you shouldn't worry about that :)  It is enough time
m
0
l
April 12, 2011 4:39:07 PM

Lanqu said:
There is about 5 mlns write operations allowed for a SSD before it will go to the trash.


Will that be 5 millions writes to the whole SSD drive totally or to each sector?
m
0
l
a c 143 G Storage
April 12, 2011 5:08:53 PM

ricno said:
As to this write wearing, will it end up to the SSD being damaged/not working if doing way to much writing (say 100 GB write/day very often) or will the performance decrease?
You will see your drive throttle it's performance down until the OP / TRIM completes it's activities to free up the necessary space. It won't be as big of an issue with the Vertex 3 and it's new firmware (better addresses the throttle issues).

Here is more on this subject... http://www.ocztechnologyforum.com/forum/showthread.php?...

ricno said:
So there is no other differences between SATA2 and SATA3 than the throughput itself? Then I can live with the SATA2 connector for the boot drive until I (at some time) upgrade to a new motherboard with only SATA3.
Your statement is true at the simplest level :na:  There are other differences but for this topic and your needs, it is the throughput. The Vertex 3 will throttle down it's performance for the SATA II drive, so it'll more closely end up reflecting the Vertex 2 performance.
m
0
l
April 12, 2011 5:15:26 PM

tecmo34 said:
The Vertex 3 will throttle down it's performance for the SATA II drive, so it'll more closely end up reflecting the Vertex 2 performance.


Thanks for your reply! With throttle down the performance, will that mean the raw throughput (down to ~300 MB/s) or other factors? That is, will there be lower amount of IOPS or higher latency or similar?
m
0
l
a c 143 G Storage
April 12, 2011 5:19:32 PM

It means the complete performance of the drive... You will begin to notice a noticeable difference in performance... example: programs taking longer than normal to open, slower read and write numbers, etc.
m
0
l
April 12, 2011 5:22:09 PM

tecmo34 said:
It means the complete performance of the drive... You will begin to notice a noticeable difference in performance... example: programs taking longer than normal to open, slower read and write numbers, etc.


Will that be of the excessive writes? I was really asking here about the SATA2 effect on the Vertex3?
m
0
l
a b G Storage
April 12, 2011 5:23:36 PM

That OCZ link would teach many a thing or two about the distinct differences between Sandforce controlled drives and all the other "traditional" controllers on the market.

Well worth the read for anyone looking to give correct advice on these controllers as not all is what it seems. Sandforce uses unallocated space quite differently than others and OP should be used more heavily to increase efficiency when larger amounts of incompressible data is used on a consistent basis.

And considering no one should be filling their drives much past 70% at any given time without first deleting something else anyways,.. why not allow the controller to keep a tighter reign on that space, right? The drive still uses the physical space.. but just makes more efficient use of it, is all.
m
0
l
a b G Storage
April 12, 2011 5:25:57 PM

ricno said:
Will that be of the excessive writes? I was really asking here about the SATA2 effect on the Vertex3?


It's called Durawrite and is part of the design of this controller. The thread that Tecmo linked you is worth the time if you're going to be using the V3. Some rules are not as extremely implemented on the newer controller but otherwise.. same-same.
m
0
l

Best solution

a c 415 G Storage
April 12, 2011 10:29:49 PM

> Is there any term for this kind of "write maximums", that is - can I look at the specification for a SSD and find how many writes it can take (as the around 5GB/day in 5 year)?

The general term is "write endurance", and it's unfortunately a hard figure to come by. One of the reasons I chose Intel was because they had quoted that figure and at the time I couldn't find a similar one for the other drives I was looking at. BTW I misquoted Intel's statement for the 160GB X25-M G2 drive - it should be 20GB/day for 5 years, not 5GB/day.


> I guess also it is kind of hard to guess how many writes a typical system does

The "SSD Toolbox" software that the Intel drives come with show the cumulative write figure which is recorded in the "SMART" data for the SSD. I know the Intel X25-M G2 drives store this statistic, but I couldn't tell you if other SSDs do.


> As to this write wearing, will it end up to the SSD being damaged/not working if doing way to much writing (say 100 GB write/day very often) or will the performance decrease?

Over time the write performance will decrease, but read performance shouldn't be significantly affected.


> So there should be no technical difference between unpartitioned space and empty logical space inside a partition?

No, not as long as you're using Windows 7 - it issues TRIM commands for deleted files so that the SSD knows it can reclaim the space. This wouldn't be true of files in the recycle bin, since it would prevent you from getting them back again. But once they're removed from the recycle bin the SSD will be able to reuse the space.


So there is no other differences between SATA2 and SATA3 than the throughput itself?

For all practical purposes, no.


Even as I do think it is very very impressive with these numbers of fantastic sequential transfers, it should be that often the 300 MB/s limit should be reached.

Yes, but you have to look at the practical side of it. The typical I/O size under Windows is around 64KBytes, which takes 21 microseconds over SATA2 and 13 microseconds over SATA 3 at 500 GByte/sec. But it also takes 65 microseconds for the drive to respond to the request (this is the "access time" for an Intel X-25M G2 drive) - that means the real-life difference is between 86 microseconds over SATA2 and 78 microseconds over SATA3 - a difference of only around 10 percent. In my experience 10% is a difference you can measure but unless you have two systems running side-by-side it's not a difference that you'll actually notice.
Share
a b G Storage
April 13, 2011 4:09:18 AM

Not gonna keep beating a dead horse here but Sandforce does in fact use unallocated space quite differently than typical controllers. Adding OP DOES allow the controller to create a larger free block pool which is swapped in and out of the logical space as required to promote increased efficiency. Using all the available physical space in formatted but free space scenarios is like comparing a drive of the original OP design(50/100/200GB models) to those that are extended capacity versions(60/120/240GB models).

The higher factory over provisioned models will always have better stamina and even slightly faster speeds. So in reality, adding extra unallocated space to extended versions acts to bring them up the higher standard of the larger factory OP original models(50/100/200GB).

The end result is a drive that has greater unallocated space which can recover quicker and maintain stamina a bit longer due to more efficient on the fly cleaning and larger reserves to reduce future throttling.

There is yet another state with Sandforce controlled drives called a "hammered state" which confuses the situation even more yet. This is the worst possible state that the controller will drop to and it can sustain it indefinitely until the writes become lighter. Usually take days to release and many hours of low activity logoff idles to recover enough for the controller to let go of the throttle.

And trust me you don't want to stay in that mode for too long because the latency goes up and the writes go down to 50% of original fresh speeds when using incompressible data(vid, pics, music, zips, rars, etc).

OP(over provisioning) can help reduce this very effectively as it reduces the likelihood of "hammered state" occurrence even when the data sets go beyond physical available space. This is possible due to faster recycling/on the fly cleaning ahead of the data streams while it plays catchup to increase the reserve back to large enough reserve state to avoid the throttle when the next data set gets written.

The Durawrite throttling process is written with clear lines drawn and trim cannot be used to recover speeds from any throttled stage.. EXCEPT.. the "hammered state" but only does so when the controller decides the data flow has slowed down and not necessarily due to Trim catching up as most believe.

Once the drive reaches the settled state throttle(often called the lifetime throttle), trim cannot recover performance back up to fresh speeds ever again unless the drive is secure erased which wipes out the Durawrites throttling ability until a new map has been reformed once more(all nand written at least once). OP can slow this process and is well worth it in the long run.

Even using the new V3 with its reduced throttling tendencies and faster GC/on the fly cleaning, I would never allocate all physically available space. OP and Sandforce are part of a good maintenance strategy and should not be underestimated.
m
0
l
April 29, 2011 12:57:31 PM


Thanks for all interesting replies. I have now bought a Vertex 3 SSD 240 GB and will mount it tonight, and will begin with a partition not covering all space.
m
0
l
May 6, 2011 7:55:36 AM

Best answer selected by ricno.
m
0
l
!