0

I very intensively use an SSD, but never fill the filesystem more than 60%.

Will the performance and endurance of the SSD increase in case with bigger Over Provisioning space ? The storage controller does not support TRIM, so there is only internal garbage collection possible.

mestia
  • 139
  • 4
  • RE: endurance, that will depend to some degree on the SSD. Can you tell us which one you're using? (A lot of the newer ones reserve some space from the full capacity to be used to replace capacity as the reads/writes wear out sections of the disk.) – HopelessN00b Dec 19 '14 at 13:41
  • The disk is Samsung 840EVO 1T. But I am asking about the impact of additional OverProvisioning, for example allocated via setmax (HPA) – mestia Dec 19 '14 at 14:44

2 Answers2

0

No, you won't see any benefit to endurance or performance from larger amounts of over provisioning ( more than 10-20% ). Note though, that over provisioning does not mean simply keeping free space within the filesystem, since adding and removing files over time will tend to write to different parts of the disk. You have to actually leave part of the disk unpartitioned so that it is never written to.

psusi
  • 3,347
  • 1
  • 17
  • 9
  • I see, I am not going to give more than 20% of OP and surely I will leave some not partitioned space or ever space in HPA. But garbage collection mechanism should be triggered at some point and more free cells should be released. So having free space can be useful anyway or am I wrong ? – mestia Dec 20 '14 at 16:26
  • @mestia, having *some* free erase blocks is helpful to the gc, yes, but having *lots* is not really any more useful than some. – psusi Dec 21 '14 at 04:03
  • this question really bothers me, I found a lot of articles which say opposite having for example 25% vs 12% of over provisioning can even double the performance on some tests. For example: http://www.anandtech.com/show/8216/samsung-ssd-850-pro-128gb-256gb-1tb-review-enter-the-3d-era/7 – mestia Jan 07 '15 at 13:34
  • @mestia, you mean like Bert's answer pointed out and I commented there that those tests are misleading? – psusi Jan 07 '15 at 13:51
0

Anandtech has a write-up on the performance of over-provisioned drives from a few different vendors. While 50% did show slight improvements over 50%, 25% resulted in similar performance and was the recommended amount to use.

Bert
  • 2,863
  • 12
  • 13
  • I've read this write up already, but it is not clear to me how exactly the test were performed, that is what my question about. Will one see such an improvement only when disk is almost full or will it be the same when there is plenty of free filesystem space and not accessible by OS overprovisioned space. – mestia Dec 20 '14 at 16:14
  • @mestia, the filesystem does not necessarily have to be full currently, only that every block in it has been written at some time in the past, and then possibly freed subsequently. Then without TRIM, the drive has to assume that all of the blocks are still in use, even if they aren't. – psusi Dec 21 '14 at 20:07
  • One needs to understand the conditions these tests were run in and how applicable they are to real life. The tests bombard the drive with writes and see how fast it can take them, and for how long. If you look at the graphs, even with no additional over provisioning, they pretty much all handle at least 30 seconds of continuous writing before slowing down at all. So unless your workload consists mostly of copying many gigabyte files to the drive from another fast ssd, you don't really need any additional sparing, certainly not 25% or more. – psusi Dec 21 '14 at 20:21