LockDown32
asked on
Samsung Enterprise SSD Drives - RAID controllers with TRIM support
I was looking at some Samsung Enterprise SSD Drives (PM863) and had a question about garbage collection. I couldn't find the answer on their web site so I called. They said there was no voice contact (support) on these drives and if you can't find the answer on our web site there isn't anything you can do. If that is true, that these drives have no support, why would anyone buy them? Does anyone know if that is true that they have not support? Does anyone know if they do garbage collection?
Along that same line.... I chatted with Intel about their Datacenter SSD drives and their knowledge was less than stellar. They said there are RAID controllers out now that support TRIM but couldn't give me any manufacturers or model numbers. Does anyone know of any?
Along that same line.... I chatted with Intel about their Datacenter SSD drives and their knowledge was less than stellar. They said there are RAID controllers out now that support TRIM but couldn't give me any manufacturers or model numbers. Does anyone know of any?
ASKER
That is what scares me. You would think that if it didn't need TRIM for what ever reason that it would be a major selling point and should be on their web site. I believe I read the same article but it would sure be nice to see that statement on their web site.
and the fact that there is no one in support to ask actually blows my mind. What kind of operation is that? I for one can't buy something that has no support and this is one prime example of why. What were they thinking......
and the fact that there is no one in support to ask actually blows my mind. What kind of operation is that? I for one can't buy something that has no support and this is one prime example of why. What were they thinking......
It may have been http://www.samsung.com/us/system/b2b/resource/2015/03/16/WHP-SSD-DATACENTER-FEB15J.pdf but it really only mentions that the advanced flash controllers in enterprise SSDs are designed to reduce write amplification to near 1 without saying how they use the extra space and the onboard RAM plus capcitors to achieve it.
ASKER
They mention garbage collection twice in the article but never say their drives to it. Very, very odd. If I had to guess those drives don't do any garbage collection. What a shame.....
Garbage collection is bad as it leads to write amplification, the enterprise SSD minimises it by using the overprovisioned space but I'm certain they do do it if they have to, otherwise the SSD would run out of space when it was only 10% full. You'll note that the article also only mentions overprovisioning as something users do with workstation SSDs but of course the enterprise SSD's overprovisioning (and GC) is hidden from view behind the advanced flash controller which they won't tell us anything about except to say it's wonderful.
ASKER
Garbage collection is bad? All the articles I have read very plainly state that if no garbage collection is done that the life of the SSD gets drastically reduced and writes, over time, will take much, much longer. One of the big plus's Intel has going for their Datacenter drives is background garbage collection. You can put them on a RAID controller and they do their on garbage collection. TRIM is not needed. One way to help is over provisioning. Samsung might do both but they sure don't mention it if they do. Either they don't do either or they need to fire their marketing department :)
Of course GC is bad, let's assume we have a tiny SSD which has 4 blocks and each block is composed of 4 x 4K pages. the OS writes in 4K sectors and for this simulation will write to each 4K sector four times. V=valid data, S=stale data, 0=clean
sectors written Aggressive GC Lazy GC
0 0000 0000 0000 0000 0000 0000 0000 0000
4 sssv 0000 0000 0000 sssv 0000 0000 0000
8 sssv sssv 0000 0000 sssv sssv 0000 0000
aggressive cleans, lazy doesn't
8 0000 0000 vv00 0000 sssv sssv 0000 0000
12 0000 0000 vvss sv00 sssv sssv sssv 0000
both clean
12 vvv0 0000 0000 0000 0000 0000 0000 vvv0
After just 12 writes using my none-too-clever OS write pattern the aggressive GC algorithm has cleaned one more block than the lazy algorithm has and it is this cleaning of blocks that wears SSDs out. GC is a harmful but necessary process.
sectors written Aggressive GC Lazy GC
0 0000 0000 0000 0000 0000 0000 0000 0000
4 sssv 0000 0000 0000 sssv 0000 0000 0000
8 sssv sssv 0000 0000 sssv sssv 0000 0000
aggressive cleans, lazy doesn't
8 0000 0000 vv00 0000 sssv sssv 0000 0000
12 0000 0000 vvss sv00 sssv sssv sssv 0000
both clean
12 vvv0 0000 0000 0000 0000 0000 0000 vvv0
After just 12 writes using my none-too-clever OS write pattern the aggressive GC algorithm has cleaned one more block than the lazy algorithm has and it is this cleaning of blocks that wears SSDs out. GC is a harmful but necessary process.
ASKER
You hit it with your last comment. What it does is bad but not doing it is worse. Guess it depends on how you look at it.
ASKER CERTIFIED SOLUTION
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
As far as RAID controllers that support TRIM I think Intel are wrong there, there are RAID controllers that support TRIM in pass-through mode but then it's not really a RAID controller if it's working in HBA mode.