How to setup a SAN...?

[Obee1]

Hi guys

 

We currently have a flat network. No VLANs, no VMs and all centralised storage is on servers and RAIDs.

 

I am considering the move to using VMs on Hyper-V and a SAN. As it stands right now I have created fairly small C partitions on the servers, so as to maximise the resource available to end users.

 

All of this works fairly reliably. We have however had a couple of issues throughout the year:

* Shared storage server went down - 2/3 days to resolve as two disks failed.

* Ran out of space when trying to regenerate boot images on our deployment server.

* 1 DC completely failed at one point and took a couple of days to rebuild.

 

I've always kind of thought that VMs are a bit of a fad if I'm honest. I don't really believe that running something virtually will provide you with the same resource as a physical device, even if your server is upscaled to cope with the extra load, that said the ability to regenerate a server in a matter of minutes definitely appeals. As does the ability to add extra storage very easily.

 

For this reason I'm considering the implementation of the above.

 

Hyper-V as I understand it is fairly straight forward although it does restrict you to just two servers per physical right?

 

With regards to SAN. I don't really know how to setup at the moment or exactly what is required. I have found a few useful resources, but not many recommendations.

Does anyone have any recommendations, or a specification of what they've had?

Anyone got a quote that they're happy to share? Our budget is extremely limited so I might not even be able to afford to implement this.

 

Any advice much appreciated.

[Jawloms]

Hyper-V as I understand it is fairly straight forward although it does restrict you to just two servers per physical right?

 

Nope. I believe the confusion here may be that a licence for a server entitles you to run two VMs on a physcial. You can actually put on as many as you like, so long as you have the grunt behind them and buy more licences.

[PotNoodleTech]

A SAN is going to overcomplicate things a lot. I really don't see the need.

 

As you are making a big jump in technology I would suggest starting simple, just getting 2 x physical servers and use local storage (preferably solid state drives in a RAID array) and putting hyperV on them, buy two copies of windows 2012 r2 datacentre which schools can pick up cheap and then you are licenced for unlimited VMs on each host, you can literally add as many Windows 2012 R2 virtual servers as you like. Set up half your VMs on server1, half your VMs on server2 and then enable replication so that each server replicates it's own VMS to the other server. That is not a cluster, but it is reasonably fault tolerant, if one server goes down the other will take over all the VMS after a short spin up period (30 seconds-2 minutes).

 

That can provide you with a really really simple to implement highly available fast system for under £5k with no nasty long term warranty/support costs or expensive network switches that you will need with most SANS.

[pantscat]

Virtualisation would, if done correctly, help to mitigate the kind of issues you've illustrated above.

 

You might not necessarily need a SAN, it would very much depend upon your requirements and expectations.

Setting up a SAN depends very much on the vendor - In my experience Nimble SANs are VERY easy to configure and get up and running, in comparison to EMC or NetApp, which are a little less plug and play!

 

You could possibly get by just by using Hyper-V replicas (and storing your VMs on the local storage in the hosts) - where VMs are replicated to other hosts, so that in the event of an issue they can be powered on from the backup host.

 

Ant

 

(edit) - Ah... @AButters beat me to it while I was making a coffee!

Edited July 18, 2016 by pantscat

[Davit2005]

Depending if you have beefy enough servers you could use datacenter licenses which gives you unlimited on the host license wise I think??

 

Our newest blades for instance with 256GB happily run 11-12 VM's.

 

Overall we have 105 Servers accross 20 Blades so could prob take 2 blade out for maintenance or outage quite easily.

 

SAN wise we use 8Gb Fibre Channel with dual everything for redundancy.

[maark]

if your budget is limited I doubt you will be able to afford a san - not sure they are totally necessary in school environment anyway.

Go to somewhere like Server Source - Europe's Leading provider of Refurbished Dell PowerEdge and PowerVault servers get 2 servers if can afford it - they often do 2 for 1s - you can then replicate between servers for recovery.

With datacenter 2012 you can have as many hyper v virtual machines you like on each physical host.

[CAWJames]

Current licensing, is Windows Standard server gives your rights to 2 VM's, if you only run hyper-v services on the host(as soon as you run other stuff on it, you will lose the right to run the second VM). Having said that that limit applies to Windows Server VM's, so you can run as many little linuxy ones as you want on Standard. Datacenter on the other hand will allow you to run as many VM's as the proc's and memory allow, the sky's the limit!

 

We run a 2 node cluster with 170ish GB of RAM and dual Xeons on both. with a PS6000 Equallogic SAN with 14TB useable. We run around 30 production VM's and they are stable and saves us a packet in having physical servers

[Oaktech]

If your budget is very limited and you want shared storage, you might want to look at using a cheap/refurb windows server stuffed with as many disks as your budget will allow and using it as an iscsi target for the machines you want to use as hosts.

 

If you go down this path, I would be looking at running the storage on a windows server with a minimum of 4 network ports that you can use as a pair of teams, and then having a reasonable gig switch (more than a 4mb packet header) with them, running off the same UPS as the servers and then having 4 ports on each host server that you can split into 2 teams - one to do storage communications and one to do rest-of-world communications.

 

This isn't a particularly elegant way of doing it and the performance might be less than you were expecting but it'll work.

 

I've just dropped 16k doing a 2 host/17tb storage/10g stoarge network project, and I consider I've still cut corners a bit because I rely on our core switch for cluster communications. Our core switch has a UPS on it, but if the ups on the core switch goes before the cluster has migrated/shut down i'm going to be in trouble!

 

I would be more inclined to stick with good, fault tolerant local storage (2 disks for OS, raid 1, as many disks as you can afford raid 5 or 6 or JBOD) and configure replication between your hosts. I'd also spalsh out on data centre - it's about £250 per host per year currently on an EES.