ESX v5.5.0 SSD Datastore Problem

[dhoggan]

Hi,

I'm hoping that someone can help with an issue I'm seeing when creating new, or migrating existing guests to SSD datastores.

It is an HP Microserver N54L that originally had a single 250 GB SSD (Sandisk) and two 500GB hybrid drives (Seagate) all running off an LSI RAID card. This has worked without issue since it was built mid last year. Recently I have added three new 250GB SSD (Samsung EVO 850) and they appear as SSD datastore. However, any attempt to create a new guest or migrate a guest from one of the existing datastores has highly variable results, including the following symptoms:

• The copy or move operation appears to complete successfully but on adding it to the inventory and attempting to start it the client dies (Windows can complain about file system corruption, blue screen or missing disk; unix guests can kernal panic), or simply loops between starting and the VMware losing screen
• Creating a new Windows guest can fail during the early-on file copy phase complaining that file xxx is missing, where xxx is different ever time.
• Creating a new unix guest (in this case a Check Point gateway) simply hangs.

The original disks have no issues.

Has any seen this? It's really frustrating as I've run out of ideas!

Thanks,

Dave

[dhoggan]

Just an update on this issue.

I've tried moving one of the Samsung EVO 850 SSDs to the host bus of the Microserver so that the RAID card is no longer in the loop. VM moving and creation works just fine and so this looks like a compatibility issue with the EVO 850s and the LSI 9240-8i RAID controller.

Before I look at returning the disks Is anyone using this configuration of drive and controller?

Thanks,

Dave

[JarryG]

I think your problem is not SSD, but 9240-8i controller. It has no on-board cache, so your are running your disks basically un-cached (ESXi does *not* do disk-caching at all). This can cause big problems (time-outs, out-of-sync, etc) especially when you want to copy data from one disk to the other, both attached to the same controller.

This problem is even more serious with SSDs, because SSD is not reading/writing data using small sectors (512B or 4kB), but much larger blocks. This is very difficult to do without cache because you have no place to store data before destination-storage is ready to accept it. I think this is not specific just for 850evo, you'd get this problem with any other ssd.

All I can recommend is to get different controller, with on-board cache (the bigger the better), with power-protection (battery or super-cap).

[dhoggan]

Hi

Yes, as the disks seem to be OK the controller is the most likely culprit. The only issue I have is that (1) I have been using SSDs on this model of controller with no issues whatsoever for over a year in another system and (2) I have been using other makes of SSD (Sandisk) on this particular controller for over half a year, again with on issues. So, it does appear to be happy with SSDs, just not this particular model.

I have read around and one option is that the SSD controller is not playing nice with the RAID card, but I can not find anything concrete on this.

Thanks,

Dave

[jakk212]

I own a N54l too and i too can reports errors after installing an SSD drive.

I've got a kingston now v300 120gb SSD that i've installed. All was fine i installed a 2012 R2 server onto the SSD and performance was great.

However soon the hypervisor began to froze - forcing me to do a power cycle of the physical server in order to get into the vsphere client (it would hang at loading inventory)

The problem only seems to occur when managing the 2012 R2 VM on the SSD.

An complete fresh install of 6.0 upd 1 did not solve anything.

I can see that according to Kingstons site the Kingston v300 series uses a custom LSI controller.