Should I over-provison my SSD?

[Black Ninja]

Hi guys,

I have do re-do my ESXI 5.5 datastore and since I am starting from scratch again I would really appreciate expert opinion on the issue so I don't have to go back and redo again:

I am going to use 2x Intel SSD DC S3700 800GB in RAID1 connected to LSI9271-8i.

Should I over-provision them (to keep consistent performance) or should I use the full capacity and count on manufacturer over-provisioning ?

Thanks

 

[m0nkey_]

should I use the full capacity

Yes. All SSDs have some kind of wear leveling, but that would depend on the make/model of your SSDs, which would be something you'd need to check the specs for.

 

[joeschmuck]

How long have these SSD's been in service and what is the current wearlevel of them? BTW, you can't use ESXi to tell you and good specs on the hard drive, well my SSD only says it's okay under ESXi 6.5. So I'd find out what that wearlevel value is and then make the assesment. Off the cuff I would use all the SSD and not overprovision or if you wanted to over provision, only do 1GB which is a lot of space. And I'm assuming that this these are also the boot drives? I did that initially for my system but due to slow speeds due to the controller I've gone away from mirrored drives. A normal drive seems to be reliable enough and I make backups of my VMs twice a week.

 

[Black Ninja]

Yes. All SSDs have some kind of wear leveling, but that would depend on the make/model of your SSDs, which would be something you'd need to check the specs for.

In mycase intel dc s3700 800GB should have 20%+ over-provisioning from the manufacturer. I know in most case (when attached to MB and with using OS with TRIM) you should just not fill them up and will be ok. But this is under RAID and no TRIM as you know.

 

[Black Ninja]

How long have these SSD's been in service and what is the current wearlevel of them? BTW, you can't use ESXi to tell you and good specs on the hard drive, well my SSD only says it's okay under ESXi 6.5. So I'd find out what that wearlevel value is and then make the assesment. Off the cuff I would use all the SSD and not overprovision or if you wanted to over provision, only do 1GB which is a lot of space. And I'm assuming that this these are also the boot drives? I did that initially for my system but due to slow speeds due to the controller I've gone away from mirrored drives. A normal drive seems to be reliable enough and I make backups of my VMs twice a week.

They only have around 15 TB written on them. I did check that yesterday when raid volume was deleted and drives were attached directly to MB to another machine.
They are not boot drives. I have a pair of S3500 120GB in RAID1 which is hosting the hypervisor only. So I can add or remove different datastores, drives without disturbing(re-install) the ESXI itself.

P.S. I do make backups for "disaster recovery" situations if they occurred, but I don't want to have to recover from disaster when drive fail, I want the machine to keep running like nothing happened while I hot swap the bad drive.:):):)

 

[joeschmuck]

TRIM has nothing to do with wearlevel so with 20% over-provisioning I'd say use them at full capacity. Too bad your RAID does not support TRIM, guess I didn't know that. It just means slower writes as the drive needs to erase blocks of data to perform the write operation.

 

[Black Ninja]

I didn't mention wear-level. Wear-level is something so far from user control it shouldn't be even mention. And is not the cause of slowness on the drive, but the lack of TRIM is. Cause it won't tell the drive about the empty block that could be put back in service. Is not just my raid card, any real hardware raid card don't support trim.

But I am hoping like you said that with 20% over-provisioning it should be fine. All I am trying to do with this 800GB raid1 data-store is to write and read tons of data without changing the performance of the drive from new after years of use.:)


P.S. Don't tell anybody, but I did use it for a DVR-Virtual Machine for my surveillance cameras outside, which I know should be done on SSD.

 

[joeschmuck]

I didn't mention wear-level. Wear-level is something so far from user control it shouldn't be even mention.

We mentioned it since it's directly related to your question about over-provisioning. It would be an indicator on if you should do it or not.

Many SSDs have Garbage Collection built in which will automatically erase blocks not used. You would need to check into your SSDs to see if they have it and if they do, you shouldn't have to worry about TRIM for these devices.

 

[Spearfoot]

Hi guys,

I have do re-do my ESXI 5.5 datastore and since I am starting from scratch again I would really appreciate expert opinion on the issue so I don't have to go back and redo again:

I am going to use 2x Intel SSD DC S3700 800GB in RAID1 connected to LSI9271-8i.

Should I over-provision them (to keep consistent performance) or should I use the full capacity and count on manufacturer over-provisioning ?

Thanks

I wouldn't worry about over-provisioning these disks. But I recommend doing a 'Secure Erase' on them. Because (according to the article below) "By doing this, you can increase the performance of frequently used SSDs for future use."

https://www.thomas-krenn.com/en/wiki/SSD_Secure_Erase

 

[joeschmuck]

I wouldn't worry about over-provisioning these disks. But I recommend doing a 'Secure Erase' on them. Because (according to the article below) "By doing this, you can increase the performance of frequently used SSDs for future use."

https://www.thomas-krenn.com/en/wiki/SSD_Secure_Erase

While this is true, it also uses up an entire erase cycle for every block and there are a finite amount of erase cycles so I would refrain from doing this. If you must, limit yourself.

 

[Spearfoot]

While this is true, it also uses up an entire erase cycle for every block and there are a finite amount of erase cycles so I would refrain from doing this. If you must, limit yourself.

Absolutely! I limit myself to doing this exactly one time -- when I first put a newly purchased, used SSD into service.

 

[joeschmuck]

Absolutely! I limit myself to doing this exactly one time -- when I first put a newly purchased, used SSD into service.

The new SSD should have been erased by the factory. But if you do not have TRIM or Garbage Collection then when a block is written to and subsequently new data needs to be written to it, an erase cycle for that block needs to occur which is what slows things down a bit. In the real world I doubt anyone would see the slowness but on benchmarks it does show up.

Well it's dinner time, Taco Monday I think. Yum!

 

[Spearfoot]

The new SSD should have been erased by the factory. But if you do not have TRIM or Garbage Collection then when a block is written to and subsequently new data needs to be written to it, an erase cycle for that block needs to occur which is what slows things down a bit. In the real world I doubt anyone would see the slowness but on benchmarks it does show up.

Well it's dinner time, Taco Monday I think. Yum!

Yes, sir. But I was speaking of a used SSD.

 

[joeschmuck]

Yes, sir. But I was speaking of a used SSD.

Ah, sorry, I looked right over the "used" part. I saw "newly purchased" and took it from there. I could see why a person may want to perform a secure erase on a used drive, it's probably a very smart idea.

BTW, the tacos were very good.

 

[Black Ninja]

I wouldn't worry about over-provisioning these disks. But I recommend doing a 'Secure Erase' on them. Because (according to the article below) "By doing this, you can increase the performance of frequently used SSDs for future use."

https://www.thomas-krenn.com/en/wiki/SSD_Secure_Erase

I always do secure erases for that reason.I think I became addicted to it.:)

 

[Black Ninja]

P.S. I seem I don't always get email when somebody reply. That's the reason I reply just now.:(:(:(

I still didn't get an email notification,but decide randomly to visit the forum and saw new reply.?!?