Which current models of SSD support trim when being used as a datastore drive in eSXi or doesn't it matter?

ESXi doesn't support TRIM.  You can get drives with advanced garbage collection, which will essentially mimic TRIM behaviour.. however those systems rely on the drive idling.  So if you have a VM running a database, or something like Sabnzdb, chances are it'll never idle and AGC won't activate.

 

I bought a Corsair Performance Pro for my rig and it's been going strong for 1.5 years now.. it has AGC, but who knows if it's actually been doing anything worthwhile

"Advanced garbage collection", that was what it was called I knew there was something I needed to look for. That Corsair drive has been superseded now, any ideas of anything current?

Not sure sorry, haven't needed to research.

Most current SSDs have much better garbage collection algorithms relative to those in past years, and will work just fine with or without OS TRIM support as long as they have a modest amount of "idle time" during the day.    Their firmware recognizes when the drive's been idle for a while, and automatically does garbage collection.    These work fine in XP, Linux, etc. without TRIM support.

 

Do you have any model recommendations gary? I've been thinking about dumping my RAID 10 acting as the datastore for a few redundant ssd's instead...

I like the Intel 335's and 500 series units, or the Crucial M500's

 

I have a bunch of the Samsung PM 840 PROs.

I like them. They are fast and the pros come with software to help migrate. (Windows)

2 of them I beat the hell out of, I mean really show no mercy, and they have been going strong since December.

 

I do not know the Intels or Crucials, Maybe someone can contrast the differences.

 

This is what I found for 240/256gb units.

 

Samsung PM840 PRO

    256GB

 

Memory Components

    MLC

 

Interface

    SATA III

 

Controller

    Samsung MDX Controller

   

Performance

 

Sustained Sequential Read

    540 MB/s

 

Sustained Sequential Write

    520 MB/s

 

4KB Random Read

    Up to 100,000 IOPS

 

4KB Random Write

    Up to 90,000 IOPS

 

MTBF

    1,500,000 hours

 

 

Intel 520 series 240gb.

 

Capacity

    240GB

 

Memory Components

    MLC

 

Interface

    SATA III

 

Controller

    SandForce

   

Performance

 

Max Sequential Read

    Up to 550 MB/s (SATAIII)

    Up to 280 MB/s (SATAII)

 

Max Sequential Write

    Up to 520 MB/s (SATAIII)

    Up to 260 MB/s (SATAII)

 

4KB Random Read

    Up to 50,000 IOPS

 

4KB Random Write

    Up to 60,000 IOPS

 

MTBF

    1,200,000 hours

 

I have used 8 of the Intel 520's in ESXi builds for customers.  Those 8 drive were in 4 different boxes in a RAID1 configuration and were being used as the ESXi datastore drives.

Just bought 2 of these for my datastore needs.

 

http://slickdeals.net/f/6275608-256gb-toshiba-q-series-sata-iii-internal-solid-state-drive-ssd-150-free-shipping

my .035 percent of a dollar?

 

Don't use a SSD as a datastore if you don't have a replacement on hand, and have your current datastores backed up. While offline-garbage collection can help. Really, it's not the end-all be-all for SSD drives.

 

I've got a SSD hosting my datastore for a ubuntu desktop, windows 7 encoding computer, and soon a Domain Controller. I however wouldn't trust any of those above to a server where I didn't have any kind of backup, whether it was VCB, Veeam, or others, no backup, no run on SSD. Oddly enough, I back all of these up to my unraid server behing hosted on the ESXi server - heh)

 

I've also got a 60gb (emergency data store) for those three computers in the house. I'm running a Corsair Force 3 256gb as a datastore, and I trust it ... zero.

 

Run SSD for the speed, but don't trust it on a ESX server. That's my 2 cents. I actually hope I'm proven wrong in the long run, but right now? I don't trust my SSDs on my servers. I trust them enough to run my laptop, desktop, HTPC1, HTPC2 ... but my servers? No Sir, don't trust them.

Run SSD for the speed, but don't trust it on a ESX server. That's my 2 cents. I actually hope I'm proven wrong in the long run, but right now? I don't trust my SSDs on my servers. I trust them enough to run my laptop, desktop, HTPC1, HTPC2 ... but my servers? No Sir, don't trust them.

I tend to agree.  I went to 300GB WD Raptor's for my DataStore drives - I had two already so I only had to buy a third for my third ESXi host.  With Thin Provisioned OS drives I still have between 110 - 180 GB free depending on ESXi host.  I do want to get a couple more Raptor's as off line backups however.

Well ....

 

"... Don't use a SSD as a datastore if you don't have a replacement on hand, and have your current datastores backed up. "  ==>  I'd say that comment is true for ANY drive.    You should ALWAYS have a current backup, whether you're using an SSD or a rotating platter drive;  and if it's an important system, you should always have a backup drive ready to use if necessary.

 

I've seen MANY OCZ units fail;  a few Plextor, Samsung, and Adata units fail; but have (knock on wood) had ZERO failures with Intel 335 or 5xx series SSDs (and have installed ~ 20 or so).    I've also never had any failures with Crucial M500 series units, but have only used about half a dozen of them, so that's a much smaller sample.    A couple years ago I used a lot of OCZs, but they've had a ridiculously high failure rate, and I wouldn't buy them today at any price.

 

But I'd have no problem using an Intel SSD as a datastore.    On the other hand, a Raptor certainly provides a "fast enough" drive at a much lower price.

 

Can you elucidate on OCZ unit failures? I mean, is there any public data available to back up your claim?

 

And yes, a raptor is a nice datastore, as long as you only use the first half of the drive ... aka short stroking it.

Don't do thin provisioned drives for a drive that could potentially have 90+ percent usage.  Don't do a thin provisioned drive where performance is important.

 

One needs to prioritize the drives on a storage device. If VM #1 needs more HD bandwidth you thick provision VM #1 on that data store. If #2 needs, it, again, thick provision that first. Doing a thick provision pretty much promises you're gonna short stroke the drive and improve on all access/transfer times for the drive.

 

For a VM you don't care about? Put it 2nd or 3rd or last. OR you could allocate resources for it. I myself tend to allocate resources on a physical vs virtual/allocation level. I know if I put the cache drive on the first 256 megs of a 1.5 or greater data store disk on my esxi server, it's gonna have much better access times, and much better throughput times, than if it was in the last 1/3rd of the disk.

 

Run SSD for the speed, but don't trust it on a ESX server. That's my 2 cents. I actually hope I'm proven wrong in the long run, but right now? I don't trust my SSDs on my servers. I trust them enough to run my laptop, desktop, HTPC1, HTPC2 ... but my servers? No Sir, don't trust them.

I tend to agree.  I went to 300GB WD Raptor's for my DataStore drives - I had two already so I only had to buy a third for my third ESXi host.  With Thin Provisioned OS drives I still have between 110 - 180 GB free depending on ESXi host.  I do want to get a couple more Raptor's as off line backups however.

Don't do thin provisioned drives for a drive that could potentially have 90+ percent usage.  Don't do a thin provisioned drive where performance is important.

 

One needs to prioritize the drives on a storage device. If VM #1 needs more HD bandwidth you thick provision VM #1 on that data store. If #2 needs, it, again, thick provision that first. Doing a thick provision pretty much promises you're gonna short stroke the drive and improve on all access/transfer times for the drive.

 

For a VM you don't care about? Put it 2nd or 3rd or last. OR you could allocate resources for it. I myself tend to allocate resources on a physical vs virtual/allocation level. I know if I put the cache drive on the first 256 megs of a 1.5 or greater data store disk on my esxi server, it's gonna have much better access times, and much better throughput times, than if it was in the last 1/3rd of the disk.

 

Run SSD for the speed, but don't trust it on a ESX server. That's my 2 cents. I actually hope I'm proven wrong in the long run, but right now? I don't trust my SSDs on my servers. I trust them enough to run my laptop, desktop, HTPC1, HTPC2 ... but my servers? No Sir, don't trust them.

I tend to agree.  I went to 300GB WD Raptor's for my DataStore drives - I had two already so I only had to buy a third for my third ESXi host.  With Thin Provisioned OS drives I still have between 110 - 180 GB free depending on ESXi host.  I do want to get a couple more Raptor's as off line backups however.

Agree.  In my case it doesn't matter as most IO is done off drives that are on a pass through controller.  It IS windows so I could improve the speed by thick provisioning the drives but performance improvements were small for me so I just didn't bother after testing out both.

Yup. Although, most of my non-stop disk access is done on VMDKS vs passthrough. My Windows 7 encoding machine utilizes a hitachi 500gb as it's D drive. This is a 360gb VMDK on a 500GiB drive. This drive reads mpgs and encodes to h264. While not a lot of traffic is passed, (6 to 22k per second depending on source) it's enough where I'm ok giving it it's own VMDK on it's own spindle. This is also done 24 hours a day (right now it's 54 shows behind in encoding)

 

The only pass-through device I have is my M1015, and that's just passed straight through to UnRaid. The other 6 ports are on the base. I do plan on getting another 1015, but that's in the future, and as Yoda said "Unknown, the future is."

 

 

as Yoda said "Unknown, the future is."

And how many of you heard that in Yoda's voice? 

*grin*

 

as Yoda said "Unknown, the future is."

And how many of you heard that in Yoda's voice? 

So, esxi doesnt support trim on datastores but it is possible to get trim support on a RDM ssd disk using lvm? Why I ask is cause I have a ubuntu guest and can't get it to work:

 

sudo hdparm -I /dev/sdb | grep -i TRIM

SG_IO: bad/missing sense data, sb[]:  70 00 05 00 00 00 00 0a 00 00 00 00 20 00 00 c0 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00

 

I modified the fstab with "discard" and set "issue_discards = 1" in etc/lvm/lvm.conf

 

sudo fstrim -v /var

fstrim: /var: FITRIM ioctl failed: Operation not supported

 

My /var parition in on lvm.

 

thoughts?