NFS versus VMFS - what is more stable and safe ?

[continuum]

In my daily work I have to deal with corrupted vmdks and datastores a lot so that means that my view is biased.
So I wanted to hear about your experience regarding the stabilty of both options.

problem	VMFS	NFS
this virtual disk needs repair - problem with snapshots	frequent	rare
vmdk is unusable after "AIOMgr_Open failed. Type 3" error	always	not seen this once
datastore partition table messed up	frequent	not seen this once
directories or vmdks are invisible after ESXi host crash	often	not seen this once
result:	unstable	stable

Is it just my biased view or is there really a significant advantage of NFS in this regard ?

Opinions welcome

[schepp]

Hey Ulli,

I only use VMFS storages on FC-LUNs (about 60 TB of VMFS storages right now) and I've never had one of those issues you mentioned in the last 3 years. So it's quite stable for me :smileywink:

Edit: I use a Fujitsu SAN.

Tim

[SG1234]

hi - please share what type of Storage it is  -- 3par/NetApp/EMC ?

~Sai Garimella

[continuum]

Cant answer that question - I have to deal with whatever my customers use

[SG1234]

the reason I ask is -- VMFS on SAN needs a robust infrastructure -- we too have seen vmdk file corruptions on 3par as the array was not capable of handling SCSI reservation/release events in time -- if the SAN network is redundant the problems should be reduced ....

NFS on the other hand seems resilient and less sensitive to glitches

helps?

~Sai Garimella

[sparrowangelste]

i've seen corruptions for VMFS on much older sans that were not able to handle the io.

once the vms were moved to vmfs volumes on newer faster sans, havent seen a problem since.

[mcowger]

problem	VMFS	NFS
this virtual disk needs repair - problem with snapshots	never	rare
vmdk is unusable after "AIOMgr_Open failed. Type 3" error	never	never
datastore partition table messed up	never	rare
directories or vmdks are invisible after ESXi host crash	never	occasionally
result:	solid	for tier 2/3 only

modified your table for what I've seen over the last 6+ years across hundreds of hosts and at least 6 different kinds of arrays.

[continuum]

wow - I find it hard to believe you never seen a vmdk that could no longer be copied or moved with "AIOMgr_Open failed. Type 3" ?

[mcowger]

Never once, and I've moves/copied tens of thousands of VMs.  Talked to some of my larger customers with tens of thousands of VMs, and the've never even heard of this.

Then again, I've also fully controlled ever environment I've ever been in, and we don't buy lowend gear.  Maybe your customers have crappy gear?

Honestly, if VMFS were as unstable as you think it is, do you really think there would be literally hundreds of thousands of successful deployments on it?

[sparrowangelste]

crappy gear is probably what makes it fail

[Josh26]

Ulli Hankeln wrote:

wow - I find it hard to believe you never seen a vmdk that could no longer be copied or moved with "AIOMgr_Open failed. Type 3" ?

Never.

I've seen my share of bizzarre and silly filesystem issues and it usually does involve VMFS(iSCSI) and an entry level switch, or one shared with the data network.

I posted a while back regarding VMFS resiliance in general, I never considered NFS would be an alternative.

I do know that the more critical VMs are placed on a VMFS backend, the more pressure there should be on VMware for it to be reliable, recoverable, and resilient to outside errors.

[elgreco81]

Hi,

Nice discussion! Thanks for posting it :smileyhappy:

I'm not as experienced as you guys but I would like to throw some thoughts to the discussion.

As I read here, scsi reservations and the way the low performance (aka crapy LOL) arrays handles them could be responsable for data corruption. Wouldn't ATS solve that? And about partition table being messed up, again , with VMFS5 and GPT, I think there's a backup of the GPT at the end of the disk that would also solve that.

I know...I'm not giving answers but quite the opposite. Also these 2 points wouldn't address all your questions but I wanted to include them in the disussion and see your feedback.

Regards,

elgreco81

[mcowger]

low performance != crappy.

To use my own gear (VNXe), its not a speed demon by any stretch (especially when configured with the minimum of 5 NL-SAS drives), but it still doesn't have issues with SCSI reservations.