Crucial M550 and Perc H310

[denisl]

I'm looking to add a pair of 256GB SSD's under a Perc H310 RAID1.
The drive is available at newegg for $89. The intended use will be as a VM datastore under ESXi 5.5

How does does garbage collection or trim get managed in this configuration?

Thanks

 

[Biren78]

How critical is the VM for latency? We tested M500's and M550's and they are not enterprise drives. If write heavy we had seg faults due to long delays in writes.

But most light usage OK.

 

[denisl]

This will be light usage. But what about garbage collection? How is that handled without being directly owned by an OS?

 

[Patriot]

This will be light usage. But what about garbage collection? How is that handled without being directly owned by an OS?

GC in the firmware is horrendous. I have seen iops drop below 10 in testing the m500, anandtech benches show the M550 and M600 to still be poor.

 

[denisl]

So how are people overcoming the challenge of adding direct attached SSD drives to ESXi?

 

[PigLover]

So how are people overcoming the challenge of adding direct attached SSD drives to ESXi?

By selecting the right drives to do it with.

There exist drives that do GREAT garbage collection and work well with ESXi, raids and other situations that don't respect TRIM. Examples are Samsung 840 pro/850 pro, Sandisk Extreme II/Extreme pro & Intel S3500/S3700.

Then there are drives with intrusive GC or that need to have frequent periods of "quiet time" in order for GC to work. Unfortunately the Crucial's (all of them, apparently) are in this second camp. You probably want to restrict using these drives to conditions where TRIM is respected like Windows/Linux single drives and Intel MB raid (with newer drivers).

 

[denisl]

Piglover - thanks for the information. It looks like of all the options you mentioned above (samsung, sandisk, intel) that the 840/850 pro's are the best priced and come in 256gb vs 240gb on the others. Any other SSD's with onboard GC to look at before I go with the 850 pro's? Also for what it's worth these are going into a home lab and hosting VM's OS and application data. User data will be going on some HGST NAS drives.

Thanks

 

[mrkrad]

I'd suggest using 30-33% OP with samsung Pro drives with LSI controllers if you want reliability!

 

[denisl]

What does 30-33% OP mean? Thanks

 

[Kristian]

Write amplification - Wikipedia, the free encyclopedia

Over-provisioning (sometimes spelled as OP, over provisioning, or overprovisioning) is the difference between the physical capacity of the flash memory and the logical capacity presented through the operating system (OS) as available for the user. During the garbage collection, wear-leveling, and bad block mapping operations on the SSD, the additional space from over-provisioning helps lower the write amplification when the controller writes to the flash memory.

 

[denisl]

So when I create a partition on the drive just set to ~180GB usable on a 256GB drive and the drive will automatically use consume blocks in the un-partitioned space if a cell is no longer reliable?

 

[Patrick]

Any other SSD's with onboard GC to look at before I go with the 850 pro's?

I have a Samsung 845DC Pro inbound. That may also be an option. OP is already done for you.

 

[Patriot]

So when I create a partition on the drive just set to ~180GB usable on a 256GB drive and the drive will automatically use consume blocks in the un-partitioned space if a cell is no longer reliable?

Yes but... that is not the main reason. SSDs rotate the Nand that is being used for the partition for wear balancing but also for garbage collection. Having more spare nand makes it easier and the routines are less intrusive performance wise.

 

[denisl]

Thanks guys very educational. I sell flash arrays and learn something new everyday.

 

[mrkrad]

Trust me the 830 and 840 PRO needs 33% OP! With the LSI megaraid controller with raid-1 or 10 it is definitely needed for 24x7x365 reliability! ESXi will throw up on the latency spikes without reducing that 512gb SSD to 337GB under duress. Especially with raid-10 .

Using RAID-1 with extent spanning (in esxi) is a better way to go IMO, but not to everyones liking. One slow drive will slow down all 6 or 8 SSD with raid-10 ! More latency is not good !