High-availability solutions for Hyper-V

5tephen · 16 Apr 2012 at 16:54

I am planning different high-availability scenarios for a customer, with a new server (either SBS Std or Server 08 Std + hosted Exchange) and I imagine it will be based around Hyper-V. The customer is suggesting having a secondary server that they can fail over to. I'm looking for a solution that will replicate the Hyper-V setup on the primary server in real-time onto a secondary stored in a separate secured location in their warehouse (LAN link, it won't be geographically separate). The solution should be as easy to fail over as possible, even to the extend that one of their (reasonably tech-minded) staff could perhaps perform the operation; this part isn't a critical requirement but would be nice to have.

I have had a quick look and am looking at DoubleTake, SteelEye, ViceVersa Pro, but would like to hear what some of you have worked with in this environment.

Sometimes I think that the cost of this can be prohibitive versus a 2-Hr response, 5-yr warranty with the likes of Dell, and redundant disks and power supplies built in. The customer will be purchasing a new server anyhow and all of this will be in place (with the exception of changing the 2-Hr to either 4-hr or NBD). Your opinions welcomed

Skidilliplop · 17 Apr 2012 at 13:03

Hyper-V can do HA, but it fails VMs from one host to the other. This relies on windows clustering services underneath so i'm fairly sure you can't do it on SBS. It also requires networked storage that supports concurrent access from multiple nodes or two storage pools with high speed synchronous replication between them.

It is usually cost prohibitive for small organisations.

Don't confuse HA and backup/DR. Invest in backup and DR first, HA second. Backup and DR are critical - HA is just a luxury that takes the headaches out of hardware failures.

5tephen · 17 Apr 2012 at 13:39

Sorry, I meant that it would be SBS hosted in Hyper-V, running on its own or as a guest of Server 2008.
After thinking a bit, even if it was something that is replicated every hour, and they manually fail over to it, they still lose an hour's work, and there's the issue of bringing back that "hour's work" (+emails if appropriate) when they bring the main server back online. So I guess the HA has to be in realtime or not at all.

Backups would be separate from all of this I understand, I have that bit covered.

All of this for a company that has about 15 client machines. Probably looking at an improved warranty to reduce the complexity of everything.

Thanks.

bluebox · 17 Apr 2012 at 13:41

The first thing to do is get a proper view of the RPO/RTO requirements.
How long can you be without the service?
How much data can you afford to lose?

If you ask those questions chances are the first answers will be no loss of service and no loss of data but when you explain that will double, triple or more the cost of the infrastructure (failover servers, shared storage, additional software costs) then you'll ge a more accurate answer.

Once you know that then you'll know what level of redundancy and DR provision you'll need.

adencool · 17 Apr 2012 at 14:36

My two ideas on this would be:

2 servers, both running hyper v server, with sbs as a vm. this then is backed up to the other machine via altaro http://www.altaro.com/hyper-v-backup/ on a daily schedule. so if the main server fails. you can simply mount the vhd (this is what altaro produces) and off you go on the second server. with at max 1 day of data lost

or

hosted exchange, and 2 servers, both running server 2008, both operating as dc,dhcp,file share, and the file share as a dfs group. this way if 1 server goes offline, no data will be lost. but you need someting like gfi max, to monitor the servers, else you wont know is one is offline

Skidilliplop · 17 Apr 2012 at 15:10

For 15 enpoints HA isn't worth it. The ROI would be unachievable within the lifecycle.

It depends exactly what they do. If it's just file and print shares then DFS is probably enough to protect data, then a small manual DNS tweak provides the failover.

Build it around how the business works and the system needs, not client desires. If you offer them HA and other flashy things of course they'll want it.

I also don't get why you would put SBS in hyper-V on top of 2008? Pretty sure 2008 STD with hyper-v allows licensing for the base (essentially the hypervisor) instance of 2008 + one virtual install. That would get you the same effect cheaper.... OR just use the 2008 install, as there's little or no need for hyper-v that I can see here.

Someone needs to sit down and do some maths here, figure out what the requirements are, how much an hour, 4 hours, a day of data could cost, what resiliency is necessary and how big the budget is to work with.

I can tell you right now, VMs rarely pay off with less than 4 VMs per host. In many cases it's a lot more. When i worked it out, ROI started to kick in at 8 VMs per host on standard 1U dual CPU boxes. Blade systems it's totally different numbers. You really do just have to sit down and do maths and work out what solution fits.

5tephen · 18 Apr 2012 at 12:25

Some good points.
Skidi - SBS Premium does indeed come with a licence to run Server 2008 as the host, but not Standard.
Also the idea of VM's was just to allow for easier replication if we were going down that route.

Anyhow after a further meeting the arrangement is to go with hosted Exchange for a start; I am going to look at fixing up the config of their existing SBS 2003 to improve a few things in the short term, and once the hosted email has settled in, look at options for the server.
They are a courier - they need to have access to the server all the time in order to print manifests and get their vehicles out the door as soon as possible. If they can't access their system to print the manifests, they said that planning alternative routes to get stuff shipped for the same delivery time promised can cost them £1000's sometimes. They have a bespoke system running off SQL Express at present.
HA for SQL Express is perhaps achievable with a few hacks but it's not a complete solution. Adencool's second suggestion sounds perfect except for the SQL side of things, and to bring SQL clustering in is just not going to be cost-effective. Your recommendation of Altaro has been noted - it looks great.

It's looking like a single Server 2008 solution, running in Hyper-V, that say replicates the VHD changes using Altaro to a NAS every 15 minutes; if there's a major failure of some kind, at least the VHD is there to mount off a spare machine and keep them going until the primary machine is sorted. Seems the most cost-effective plan for a business of this size.
(Backups again would be running to separate, rotated disks)

Skidilliplop · 18 Apr 2012 at 16:39

For something that lightweight why not host the lot? Online backups and DR all taken care of under the service and as most hosted servers these days are VMs running in a redundant architecture you solve availability issues too.
You also have the fallback of SLAs so if you do suffer loss of earnings by prolonged outages you get compensated for it.

If you look at the total cost of the separate HA and DR plans, you'll probably find the money you save combining the two equates to a couple of days loss of earnings... Making the economics questionable. As I said before, Maths. Speculating is useless without the actual hard numbers. With those the decision becomes black and white.