SSDs and Virtualization

[Ipreferspam]

I am very interested in purchasing two or three SSDs for use in a server. I am concerned about the reliability of SSDs in my intended application use, however, as I will be running VMWare ESXi off of a RAID0 array of these drives.

On top of ESXi, I will be running multiple Windows Server 2008 R2 and Ubuntu 9.10 server virtual machines. One of these will be a busy web server and another a busy database server.

I am currently considering either the Intel X-25E 32 GB and the OWC Mercury Extreme SSD 100 GB (or some other SSD using the SandForce SF-1500 controller).

The Intel due to it being SLC and therefore more reliable for server use, the OWC since it has a 5 year warranty and is guaranteed to not have a degradation in performance over those 5 years (due to lots of reserved space and the SandForce wear leveling, etc.).

However, I know that in RAID0 that I would be unable to use TRIM for either drive. And since these drives will have virtual servers running on them, there will be no possibility of a TRIM command ever being sent to the drives. With that said, are they well suited for a usage pattern similar to what I intend to use them for?

In other words, is the internal garbage collection of the drives (either SLC SSDs or a SandForce SSD) up to the task of maintaining the performance level of the drives if they are used in a busy virtualized environment?

Thank you!

 

[Hacp]

I am very interested in purchasing two or three SSDs for use in a server. I am concerned about the reliability of SSDs in my intended application use, however, as I will be running VMWare ESXi off of a RAID0 array of these drives.

On top of ESXi, I will be running multiple Windows Server 2008 R2 and Ubuntu 9.10 server virtual machines. One of these will be a busy web server and another a busy database server.

I am currently considering either the Intel X-25E 32 GB and the OWC Mercury Extreme SSD 100 GB (or some other SSD using the SandForce SF-1500 controller).

The Intel due to it being SLC and therefore more reliable for server use, the OWC since it has a 5 year warranty and is guaranteed to not have a degradation in performance over those 5 years (due to lots of reserved space and the SandForce wear leveling, etc.).

However, I know that in RAID0 that I would be unable to use TRIM for either drive. And since these drives will have virtual servers running on them, there will be no possibility of a TRIM command ever being sent to the drives. With that said, are they well suited for a usage pattern similar to what I intend to use them for?

In other words, is the internal garbage collection of the drives (either SLC SSDs or a SandForce SSD) up to the task of maintaining the performance level of the drives if they are used in a busy virtualized environment?

Thank you!

I am not too sure about virtualization but I do know that one of the big factors that adds to performance degradation is the lack of free space. Thus, you want as much free space as possible. 32gb just won't cut it.

Btw, wear leveling and performance degradation do not have a direct relationship. Wear leveling is the act of evening the number of writes in each block so they will fail at the same time. Performance degradiation happens when your drive fills up with too much valid data and the algorithms that erase invalid data aren't efficient enough anymore.

Trim solves degradation by telling to drive to wipe each 4k block individually. You can also solve the problem by keeping a bunch of space in your drive unformatted, thus giving the controller more room to breathe.

I think you'll be much better off with the sandforce than the intel because it basically has more space.

Here is an excellent paper that is easy to read that explains a lot of this stuff.
http://portal.acm.org/citation.cfm?...CFID=80951719&CFTOKEN=29566532&ret=1#Fulltext

 

[Ipreferspam]

I read somewhere a while ago that one of the reasons the Intel X-25E was such a great drive was that it had a large amount of reserved space (I am trying to find this source again but have as yet been unable to find out exactly how much space is reserved in the X-25E 32GB).

I know that the OWC Sandforce 100GB drive has 28 GB reserved internally, giving the drive more than 20% of reserved space.

I'll limit the size of my partition to provide even more unused space if that will help.

I guess my main question still remains: is it likely that I won't experience a major degradation in performance over five years on either the X-25E or a SandForce SF-1500 SSD if sufficient reserve space is left? I'm really worried about this since TRIM use will be out of the question.

Note that I'd use at least 3 of the X-25Es and that would be sufficient space for me at the moment.

Thank you, Hacp, for the link to the paper about SSD write amplification analysis. I am going to read it now.

 

[Hacp]

I read somewhere a while ago that one of the reasons the Intel X-25E was such a great drive was that it had a large amount of reserved space (I am trying to find this source again but have as yet been unable to find out exactly how much space is reserved in the X-25E 32GB).

I know that the OWC Sandforce 100GB drive has 28 GB reserved internally, giving the drive more than 20% of reserved space.

I'll limit the size of my partition to provide even more unused space if that will help.

I guess my main question still remains: is it likely that I won't experience a major degradation in performance over five years on either the X-25E or a SandForce SF-1500 SSD if sufficient reserve space is left? I'm really worried about this since TRIM use will be out of the question.

Note that I'd use at least 3 of the X-25Es and that would be sufficient space for me at the moment.

Thank you, Hacp, for the link to the paper about SSD write amplification analysis. I am going to read it now.

Performance won't suffer much if enough space is reserved and the garbage collecting algorithm is efficient. Trim won't help you if your drive is filled to the brink with valid data. You take a 32gb drive and fill it with 31gb of data. Trimmed or not, the drive will perform like crap.

 

[Ipreferspam]

I've sent a similar question to OWC's pre-sale customer support department and I'll post here what they say.

Additionally, I've found a forum post where somebody called SandForce and asked about TRIM and said:
"Just got off the phone with a 'Consumer Sandforce Rep' ... seems like the TRIM command is basically pointless with these next-gen drives. From how he explained it the way the drive wear-levels itself is kind-of like internal trim...but trim is still compatible, the drives are just so fast you'll never notice any degradation -ever."

(source: http://www.ocztechnologyforum.com/f...ys-think-Pretty-good-huh!&p=489950&viewfull=1)

If OWC replies with something similar and they guarantee that performance over their 5 year warranty, I'll definitely go with their drive as it is $10 more than the X-32E, while offering very similar performance and nearly 3x the capacity.

 

[Ipreferspam]

OWC got back to me and told me that their drives are perfect for my intended use, so I ordered 2 of their 100 GB SandForce SF-1500 SSDs.

Does anybody have any idea how I can try to put the drives in a used state when running ESXi on them with some virtual machines? I'd like to do some basic testing before turning it into a production server.

If anyone can think of any tests, I'd love to write something up to help the community.

 

[Voo]

Well.. just fill the drive up with some data, that should do the trick fairly easily

Maybe test the drives empty with only ESXi + VMs and then fill the drives up and measure the difference in throughput.

 

[Ipreferspam]

"The more spare-area we ship with, the longer our performance will remain at its peak level. But again, you have to pay the piper at some point.

Intel ships its X25-M with 7.5 - 8% more area than is actually reported to the OS. The more expensive enterprise version ships with the same amount of flash, but even more spare area. Random writes all over the drive are more likely in a server environment so Intel keeps more of the flash on the X25-E as spare area. You’re able to do this yourself if you own an X25-M; simply perform a secure erase and immediately partition the drive smaller than its actual capacity. The controller will use the unpartitioned space as spare area."
(From: http://www.anandtech.com/storage/showdoc.aspx?i=3531&p=9)

Hopefully the >20% free space and the internal garbage collection on the SandForce SF-1500 controller SSDs will help keep the performance numbers high over the lifetime of the drives.

Following Voo's advice, I plan on testing the drives empty with only ESXi + VMs in RAID0 and with a single SSD. I will then fill the drives and measure the difference in input, again in RAID0 and with a single SSD.

I understand that certain tests are not very good at measuring SSD speeds. What benchmarks are recommended to give me an idea of the performance I'll see form a server point of view?