What are the benefits of raising the SSD over-provisioning and why enterprise has 28%

[Morthawt]

http://www.kingston.com/en/ssd/overprovisioning

On that link they show client systems using 7% OP on their SSD's, while enterprise uses 28% OP on their SSD's. What is the benefit of raising the OP, why do enterprises have 28% OP and would there be any benefits to a client (typical user) system using 28% OP?

[kalniel]

What is the benefit of raising the OP, why do enterprises have 28% OP

Means the drive can carry out more write operations before end of life

and would there be any benefits to a client (typical user) system using 28% OP?

No.

[Morthawt]

It appears from graphs on that page that the actual performance also improves, the more OP you add. It seems to me it comes down to the reduction in usable space, because the more I look into this the more it seems like a good thing to do. Unless you need the space its self.

[kalniel]

It appears from graphs on that page that the actual performance also improves, the more OP you add.

That would be odd. It shouldn't matter whether spare flash was labelled as overprovisioning or just spare space, it should still contribute equally to the performance effect.

If you were to say the amount of spare space affects performance, then I'd agree. But then, removing free space from one area to add it to over-provisioning doesn't gain you anything.

[Morthawt]

That would be odd. It shouldn't matter whether spare flash was labelled as overprovisioning or just spare space, it should still contribute equally to the performance effect.

If you were to say the amount of spare space affects performance, then I'd agree. But then, removing free space from one area to add it to over-provisioning doesn't gain you anything.

"2. Why is OP important?
OP has a direct effect on the SSD's random performance as the drive is filled with data. Guaranteeing free space to accomplish
the NAND management tasks (GC, wear-leveling, bad block management) means the SSD does not have to waste time
preparing space on demand, a process that requires additional time as data is copied, erased and recopied. An added benefit
is that OP makes all of the SSD maintenance procedures more efficient, reducing the Write Amplification Factor (WAF) by
ensuring there’s room to work, which improves the SSD’s lifetime.
The below table is from internal test results that show the benefits of OP in performance and lifetime."



http://www.samsung.com/global/busine...ovisioning.pdf

[kalniel]

So essentially, there's no difference between free space and OP space. Removing free space to to create OP space is pointless.

[Agent]

http://forums.hexus.net/storage/3146...ing-ssd-2.html

The reasons for performance gains are very clear and easy to understand. As Kal says, the 28% OP in enterprise is about reliability, not performance directly.

For consumer drives, you don't need a large OP, as they are not really subject to the circumstances that would require it on a regular basis.

[Morthawt]

Maybe not performance directly, but it does positively impact performance and the, (forgot the word but) the smoothness/continuation/reliability or what ever, of the performance without dips in performance. That I have seen on a review. They said the Samsung 850 Pro with it's 7% OP was on par with some other big name drive that had 12% OP and when the 850 was boosted to 12% it clearly smashed the other drive's relentless performance (or what ever the damn word is I don't recall right now ).

edit
Consistency.. http://www.anandtech.com/show/8216/s...r-the-3d-era/7

[Agent]

Yes, I don't think anyone would debate that, but overprovisioning amounts really is situational.

If you look at the test Anand does to see how the drive copes, it is quite brutal. It's not something a home user is likely to come across.

To test IO consistency, we fill a secure erased SSD with sequential data to ensure that all user accessible LBAs have data associated with them. Next we kick off a 4KB random write workload across all LBAs at a queue depth of 32 using incompressible data. The test is run for just over half an hour and we record instantaneous IOPS every second.

So sure, it's a test to show it off, but you're not likely to come across this situation unless you have an extremely unusual work pattern. Not only that, but even with the IOPS dropping, you get an amount that is still much higher than a spindle drive.

I'm not saying the data isn't useful - just make sure you keep perspective on what it represents.

[Morthawt]

Well so far from what I have seen, I am likely to set it to at least 14% just to add the extra consistency/performance but not take up as much space as 28% would, because that is just far too much space to just vanish away. Either way after seeing all kinds of data, I can guarantee that I will be raising it above 7%.