File servers and virtualisation.

[localzuk]

I'm planning our 'mid way' migration to a virtualised solution at our school, and as it stands I have it planned as such:


3 x Dell R610 servers, each with 2 x quad core xeon processors and 24 - 32GB RAM.
1 x Dell 1950 server, connected to an MD1000 DAS storage array, running Windows Storage Server 2008 R2.


Now, when it comes to file serving, what is the best route to take? Am I right thinking that it would be best to simply set up one drive array on the WSS server to be an SMB host, with the files shared directly?

Would this make sense, as otherwise, if the file server was virtualised, the data being served would end up following this path:- DAS->Virtual Box->Client, meaning that double the traffic is traversing the box as necessary?

[dhicks]

Would this make sense, as otherwise, if the file server was virtualised, the data being served would end up following this path:- DAS->Virtual Box->Client, meaning that double the traffic is traversing the box as necessary?

When you say "virtualised", do you mean "virtualised, and running on one of the Dell R3610 hosts"? I'd avoid that and keep the file server and the data for its files on the one physical machine, although while I was at it I'd make the Dell 1950 a virtual host and just assign a VM machine running Windows Storage Server to it, then you can move the virtual machine around later if you wish. I don't know if the virtual machine system you are using allows you to assign underlying storage block devices from the underlying OS directly to a virtual machine, but if I was setting this up with Xen I'd probably simply assign the WSS VM the RAID block device provided by the MD1000.

[localzuk]

I'd be using Hyper V as the virtualisation system. How would virtualising a WSS host onto it help, as it needs access to that physically connected device via the controller card? It couldn't be moved about, as that server is the only one connected to the MD1000.

[dhicks]

How would virtualising a WSS host onto it help, as it needs access to that physically connected device via the controller card?

I assume the Hyper V host would handle the driver and so forth for the RAID card, passing the virtual machine a generic block of storage. If, in the future, you move servers around (or if you have to replace that server in case of failure, etc) you can simply move the VM to another physical server, attach the storage and away you go. I don't know how well Hyper V handles passing storage through to virtual machines, though - if it requires you to have a filesystem on the RAID array and then put a file containing a VM disk image on that file system, that might be a few too many steps of abstraction to make it worth the trouble.

[CyberNerd]

I assume the Hyper V host would handle the driver and so forth for the RAID card, passing the virtual machine a generic block of storage. If, in the future, you move servers around (or if you have to replace that server in case of failure, etc) you can simply move the VM to another physical server, attach the storage and away you go. I don't know how well Hyper V handles passing storage through to virtual machines, though - if it requires you to have a filesystem on the RAID array and then put a file containing a VM disk image on that file system, that might be a few too many steps of abstraction to make it worth the trouble.

This is essentially how we do it. We map the raw filesystem to the (VMWare) virtualhosts, so if we need to move a servers or add clustered failover then the files are all in the same place.

[plexer]

The MD1000 is direct attached storage and has to be connected to something physical which I assume could be the R610 themselves but it looks like the original plan means connecting the MD1000 to the 1950 which then dishes out disk space to the hyper-v host machines?

Ben

[localzuk]

The MD1000 is direct attached storage and has to be connected to something physical which I assume could be the R610 themselves but it looks like the original plan means connecting the MD1000 to the 1950 which then dishes out disk space to the hyper-v host machines?

Ben

Yup, that's the plan. The MD1000 would become a sort of 'unified storage' device, via a 1950, providing the shared storage for the R610s and also providing storage for the regular file storage.

[plexer]

Are you panning on replicating that anywhere? is it single controller interface to the 1950?

Ben

[localzuk]

Are you panning on replicating that anywhere? is it single controller interface to the 1950?

Ben

No replication, just like there isn't any at present. This is stage 1 of moving to a proper virtualisation solution - in 2 years, the equipment is all due for replacement and at that point I will be pushing to get 2 x SAN devices, and dual core switches to handle it all properly.

Currently connected by a single interface, yes.

[AButters]

My question is - which of the kit do you need to buy and which do you already own?

I'd be wary of spending any money at all if you're going to replace the whole lot with something far more suitable in two years.

If you already have the kit, then your plan should work.

[localzuk]

We own all of it, bar an extra processor for each R610 and a bit of extra RAM. Total cost about £3k.

There's also a necessity involved about wanting to migrate to Server 2008 R2 and Windows 7 on site, with us not doing an in place upgrade due to the existing network set up being a mess due to various BSF related reasons.