VHD on CIFS via SMB 3.0

[enuro12]

So i know this topic has been discussed. I found the bug report, and the numerous posts saying it is fixed. So maybe i'm just doing something wrong.

MY Freenas Box
8gb DDR3
Dual Core Intel Something
4x2TB WD Red
Gigabit nic
8gb sandisk flash drive

MY Hyper V Box
Server 2012 (Not R2)
SSD for os
4tb Mirror to play with
Dual 6212 Processors
72GB ram
Gigabit Nic

I built a freenas box just to test this. It is a RAID 10 of 2tb WD Red.

All the settings are default.

After setting up the share, i copied a VHD over to the share. When i attach it i get the classic message saying it cannot be done.

Digging around i noticed SMB 2.0 was the default max. I changed that to 3.0_00.
Same error.
If i change the minimum to 3.0, free nas is no longer accessible. Only when minimum is 2.x or below or 3.0_00 will it show freenas as being part of the network.

As you can see Server 2012 sees the SMB 3.0 protocol.

 

[zambanini]

I would not bet, that SMB direct will ever work on an OSS system.

You might use iscsi.

 

[anodos]

I believe in the samba3 world smb3 is a superset of smb2. Setting min protocol to smb3 breaks stuff. Macs also tend to negotiate NT1 and SMB2 connections and so setting min protocol above NT1 would probably break samba for Mac clients. In general it's best not to fiddle with min protocol.

Max protocol is set to smb2 because smb3 is buggy and not recommended at this point. As mentioned above try iscsi instead, but having seen your hardware specs I wouldn't use that box for anything other than testing.

 

[enuro12]

I had a feeling iSCSI was the only solution here. I'll probably test out Nas4Free on the other identical system i built.

I do find your statement about not using it for anything other than testing interesting. I don't see how the little test box should not be expected to perform at top speed? I'm not using dedup so my ram requirements are pretty low, embedded os and Raid10. What gives?

 

[jgreco]

For about the same reason a VW Bug doesn't perform like a Ferrari. Despite both being cars and having four wheels, one will perform at top speed and get stares from pedestrians while the other will only do the basics and only attract the attention of kids playing "punch buggy."

 

[enuro12]

I'm unsure why you guys want to attack a test environment. I clearly stated it was such. If you think i need a ZIL and 128GB of ram to test SMB3.0 on Hyper V then i've clearly asked the wrong community for enlightenment.

So how do you get me wanting to test SMB3 as me expecting a $300 test box to perform as a Ferrari? Is there something majorly wrong with my text environment? It does 113mb read/write just fine. I cant see how i could expect more from a gigabit connection.

Lastly if you're simply mistakenly thinking Hyper V shouldn't be used on SMB3 think again. That's one of the key features of 2012 R2.

 

[anodos]

I'm unsure why you guys want to attack a test environment. I clearly stated it was such. If you think i need a ZIL and 128GB of ram to test SMB3.0 on Hyper V then i've clearly asked the wrong community for enlightenment.

So how do you get me wanting to test SMB3 as me expecting a $300 test box to perform as a Ferrari? Is there something majorly wrong with my text environment? It does 113mb read/write just fine. I cant see how i could expect more from a gigabit connection.

Lastly if you're simply mistakenly thinking Hyper V shouldn't be used on SMB3 think again. That's one of the key features of 2012 R2.

I was just thinking "dual core intel something" may not have ECC memory, and 8GB RAM is not much for ZFS.

 

[mjws00]

The question is does smb3.0 as implemented by Samba support vhd and vhdx adequately for hyper-v in 2012 and the goodies in R2?

Truthfully, to get to bleeding edge samba I'd look towards linux, and roll your own. If it doesn't work on the samba nightlies, we don't have a hope on an appliance where some version lag is expected for stability.

If you get it working nicely, pipe up. Hard to say when Samba will implement some of the things MS has done. Some of the server side stuff works. Others not so much.

Good luck.

 

[anodos]

I think there is some support for hyper-v in the 4.2 branch of samba, which is still in RC.
https://marc.info/?l=samba-technical&m=141337475518640&w=2

 

[enuro12]

Thanks for the replies. I mistakenly thought MS took a SMB 3 standard and didn't re-create the wheel. I got excited about a few articles i read about SMB3 and was dying to give it a try.

I saw it was functioning in 9.2.1.7 with a bug in 9.2.1.8 marked as fixed in 9.3 Stable, so i thought it was just me overlooking something painfully simple.

 

[cyberjock]

So how do you get me wanting to test SMB3 as me expecting a $300 test box to perform as a Ferrari? Is there something majorly wrong with my text environment? It does 113mb read/write just fine. I cant see how i could expect more from a gigabit connection.

I'll tell you when I admonish test environments.. when they don't reflect reality.

Sorry, but high throughput is only 1 aspect of a server, and considering all the other stuff, your box might be able to do full throughput (which doesn't require much hardware) but I have no doubt your system would fall flat on its face if you tried to do iozone tests.

You might think your "test build" is just fine for testing, but I'd disagree. The hardware doesn't reflect anything you'd consider using in production and as a result ZFS's internal behavior wouldn't reflect reality either. So anything you think you learn about ZFS with that "test system" is suspect, and may lead you to buy hardware that is even less appropriate for a particular situation because you've seen how ZFS mismanages itself when improperly administered.

 

[enuro12]

considering the 'test' was for a protocol it accomplished exactly what it was built for. Making any other claim about an insufficient build to determine that is just silly. Would it fail iozone test? Maybe, who cares... It wasn't built for that, and to say the lack of ram or any other under built aspect is simply arguing an invalid point.

Would my system be pointless to test any throughput ? Sure. I never made any claim or suggestion about any performance beyond SMB 3.0 compatibility with FreeNAS 9.3

 

[anodos]

To be a bit pedantic (because it's what you do on the internet) I believe the feature set you are looking for was introduced in SMB3.02 (Windows 8.1 / Server 2012R2), which is still very much a work-in-progress for samba. Specifically, for VHD functionality windows will try to set resiliency on the VHD file and turn off compression otherwise file creation will fail. This means that samba needs to be able to handle SMB3.02 resiliency requests as well as fake windows into thinking compression is turned off.
These will probably eventually take the form of a VFS module. This probably means no support for VHD in Samba 4.1.x. You will have to download the current 4.2 source and compile with the correct compile-time options to get the proper vfs module. Samba releases tend to happen whenever the samba team determines it's in a good enough state for production use. Maybe we'll get VHD support by the time FreeNAS 10 comes out. :)

 

[cyberjock]

Just an FYI... I saw a discussion that Samba 4.1.5 is out (and we're not even using 4.1.4 yet) so an update for Samba on FreeNAS should be expected in the future. What that timeframe is is anyone's guess, but I wouldn't necessarily bank on an update to Samba 4.2 (which is a fairly significant jump from 4.1.x) until FreeNAS 10.x.

 

[jgreco]

considering the 'test' was for a protocol it accomplished exactly what it was built for. Making any other claim about an insufficient build to determine that is just silly. Would it fail iozone test? Maybe, who cares... It wasn't built for that, and to say the lack of ram or any other under built aspect is simply arguing an invalid point.

Would my system be pointless to test any throughput ? Sure. I never made any claim or suggestion about any performance beyond SMB 3.0 compatibility with FreeNAS 9.3

You asked, and I quote,

I do find your statement about not using it for anything other than testing interesting. I don't see how the little test box should not be expected to perform at top speed?

Which some of us answered, which you then characterized as an attack:

I'm unsure why you guys want to attack a test environment.

and made some other uncalled-for comments. Free clue: if you ask a question, and get an answer, the answer is probably not an "attack" just because you don't like the answer.

The problem with ZFS and block storage is that fragmentation effects quickly destroy performance, and that what you'll quickly discover is that performance falls off at a rapid rate if you've failed to resource the box. Even for testing, this can invalidate "testing" depending on what you are testing it for. It's certainly not good for production traffic, which was the point that had been made, and then you found that interesting and asked why.