Disaster Recovery and vmware

ashrobbo

ashrobbo

ashrobbo

Associate
Joined
3 May 2009
Posts
805
Hi All,

I am trying to talk my boss into looking at virtualisation. Managed to Load ESXi onto an ML115 and setup a few machines, all is fine and dandy, works but But hes not convinced its worth the risk loading several machines onto the same bit of kit due to 'single point of failure' which i kind of agree with to an extent.

Bit of background as to why im looking at it: we have around 28 servers, most perform 1 task only, usually a simple application, file server, print server, etc.

With all of these servers you can imagine the power usage is insane (mostly dell equipment anything from powervault nas, to poweredge 750/850/1850/2500/2950's) and reducing the power usage by 50% can only make the IT dept look good right?

Now my argument is that the utilisation on these servers are ridiculously low (some barely hit 10% cpu, and 30-40% mem utlization max)

I have suggested vmware to make the most of our hardware, but how do i get around the 'single point of failure' card he keeps drawing?

I think i could reduce our servers by atleast half and still keeping the mission critical DC/Exchange/SQL etc on seperate boxes.

We dont use any types of shared storage (few NAS boxes for backups), purely servers loaded with disks hosting whatever data/application required.

Am i missing something?

Ash
 
You need to setup a SAN, it's not expensive. Once done you can spend more money on license upgrades and get VM migration.

Edit: quick and dirty is just backup VM disk images (daily snapshot) Recovery is relatively quick but hardware failure still takes out a big chunk of infra. Still, I'd the wrong bit fails in day to day on separate servers you still have this problem.
 
Last edited:
You wouldn't virtualise onto one server for a start.

Vmware is pretty pointless without a SAN, so get used to the idea of investing in a decent SAN if you don't already have one. Single points of failure here would be eliminated by RAID, dual controllers, dual PSUs, multiple switches and multiple HBAs (or NICs for iSCSI)

With the shared storage, VMWare high availability will restart the VMs from a failed host on other hosts in the cluster. Fault tolerance keeps 2 machines running so a host can fail without any interruption - very neat indeed.

Obviously on the server side, youll have redundant PSUs, redundant disks(really only used to boot ESX), redundant fans and the usual stuff

You need to start by evaluating the current infrastructure, what sort of resource you need, what hardware you can reuse etc.

But none of this comes free, the licensing to do it will set you back thousands, as will the SAN. But you'll end up with an infrastructure that's many times more resilient than standalone servers.

I did a build for someone at the weekend - 2 hosts connected to an IBM FC SAN, running about 30 VMs. He wanted to test HA, so we literally pulled the mains out of one of the hosts - obviously all the machines running on it went down. 4 minutes later, they were back. How long would it take you to rebuild a physical server?
 
should i be looking at the fibre route or gigabit ethernet?

Cost is ALWAYS an issue where I Work, so it needs to be feasible to save the company £££ in the long run.

if the SAN Dies, then what? (cause i know thatll be the response)
 
just saw post above. a bit more info.

Server all run Dual PSU and some sort of RAID protection, we backup each server on a standard rotation to NAS.

If any of the server's died, we could probably restore within a few hours, providing we had the hardware.

My boss's opinion is that virtualization is just a 'buzzword' where as I dont agree. Thanks for responses thus far!
 
Either or, iSCSI has come a long way and costs less per host, just keep it on it's own network with good switches.

At the end of the day, once you eliminate all the redundancy in a SAN, it's just a box. But you can get 2 and have them replicating if you're willing to shell out.

However, a word of warning - do NOT try to do this on a budget with open source software and cheap hardware. False economy.

There are plenty savings to be made in the long run - management, electricity, DR costs etc etc, but they need to be willing to make a decent investment upfront

Virtualisation is definitely not a buzzword, he needs to get with the 21st century! Where I work, we have about 70VMs running on 3 servers. Management is a breeze, deployment, disaster recovery and performance are superb.

The time to do it is when your existing kit is end of life and you can justfiy the expense directly against hardware costs saved
 
Last edited:
Virtualisation can be done on a shoestring, or you can invest heavily.... I don't think you can do HA with the free version of VMWare... You could put a bunch of non-critical servers on an ESXi box - that's how we started.

Also a SAN could be viewed as a single point of failure, but to have the chassis fail would be a rare occurrence.. you're more likely to have a disk fail - being as the SAN will be RAID, this isn't a problem - they normally have dual power supplies etc too.. so no problem there.

We invested ~70k in our VMWare infrastructure.. 3*ESX nodes, 1*vcenter server, 1*iSCSI SAN.. so we get High Availability.. that means if a node fails in the cluster, all the VMs fail over to an alternative node. The VMDK file for the VM is stored on the SAN.

Our SAN now runs a file server cluster, our mail server,web servers.. etc..

The things we have kept physical are DC's and Databases.

You need to do a lot of planning to get a proper solution - make sure you have sufficient capacity in your virtual infrastructure - IOPS/CPU/RAM.

Give me a shout if you have any Q's.. happy to help.

-Edit

Wow loads of responses since I started writing - mostly saying what I said.

Good point re the SAN network - we have two gigabit switches connecting our SAN to the VMWare infrastructure..

The cost can be prohibitive.. however, if you have a 5yr replacement program, to replace 30 servers in 5 years time will cost you an arm and a leg - VMWare could reduce your overall maintenance cost, as well as your power cost. As mentioned, your infrastructure will be far more resilient too.
 
Last edited:
The n you need to give him a quote for single bs replicAted SAN solutions, and explain the difference in costs and let the beancounters decide.
 
Being fairly new in this area, can anybody reccomend any HP kit (or dell) that I can begin looking at? including licences?
i presume i would need something along the lines of

2 x servers
1 x SAN (or server with openfiler as an iscsi target?)
Switches (any particular kind? use procurve atm but majority are the older models)

We are a Small/Medium Business, i would like to budget say £5k? hopefully consolidating 6-7 physical servers into 2 boxes?

Lets forget about High Availability and replication, and i can stick with standard backups, id imagine my boss is just being difficult tbh.
 
sldsmkd said:
The n you need to give him a quote for single bs replicAted SAN solutions, and explain the difference in costs and let the beancounters decide.
Click to expand...

Thing is - the SAN is normally all redundant anyway.. the chances of losing the actual chassis are small. Probably only caused by physical destruction (fire). If you had a fire in your server room, the same result occurs whether your infrastructure is virtual or physical.
 
Ashleyrobinson said:
Being fairly new in this area, can anybody reccomend any HP kit (or dell) that I can begin looking at? including licences?
i presume i would need something along the lines of

2 x servers
1 x SAN (or server with openfiler as an iscsi target?)
Switches (any particular kind? use procurve atm but majority are the older models)

We are a Small/Medium Business, i would like to budget say £5k? hopefully consolidating 6-7 physical servers into 2 boxes?

Lets forget about High Availability and replication, and i can stick with standard backups, id imagine my boss is just being difficult tbh.
Click to expand...

Forget it.£5k won't get you past licensing, let alone a SAN.
 
5k will buy you one decent host, so forget it if that's the budget and you dont want to risk a single host failure
 
isn't it technically possible to have 3 servers, one being loaded with openfiler as an iscsi target, other 2 loaded with ESXi and vm machines stored on Openfiler?

Or is it going to be totally gash at this level?

as stated the utlization isnt huge.
 
Just no to open filer. Great for testing or the like, but what do you do when it falls over?

If you cant do it properly then don't do it at all, its not worth your job or reputation.

Wit 5 out of 28 you're not going to see any real benefit anyway. I'd be looking at putting forward a business to spend 20-30k and virtualise the lot. Gives you a standardised infrastructure, brand new supported servers, a more resilient infrastructure and big energy savings
 
Last edited:
Thanks for the info chaps, maybe I will just give it a knock on the head and let those in power call the shots, at least Ive had a look and had a go!!

Many thanks for all the responses.

Maybe i will revisit this another time when the budget stretches higher.
 
Get a good, well reasoned business case together to virtualise the lot.

Worst case, they don't go for it but you impress your boss. Best case, they go for it amd you get a really good project under your belt.

Show them you can think big and not just about the technology. Look at costs and savings of several options, talk about benefits and risk, tie it in with any applicable company policies or strategies.

Could do wonders for your career :)
 
iaind said:
Get a good, well reasoned business case together to virtualise the lot.

Worst case, they don't go for it but you impress your boss. Best case, they go for it amd you get a really good project under your belt.

Show them you can think big and not just about the technology. Look at costs and savings of several options, talk about benefits and risk, tie it in with any applicable company policies or strategies.

Could do wonders for your career :)
Click to expand...

+1 - don't give up. Cost it up and put a proposal together and let them decide.
 
For the kind of budget you have i'd be looking into Citrix Xen server also, you can get the hypervisor for free with certain flavours/features. As a guide, for a pilot, i bought 4 dell servers with dual processors and 40gb ram, an 8 cpu vmware licence, windows datacentre licence for each box (helps with licencing virtual hosts) for around £60k (included 5yrs hardware and vmware support).
 
To be honest if you simply wanted to reduce running costs and get some consolidation you could get your current 28 machines down to something like 10 ESXi hosts while still not putting all your eggs in one basket and it wouldn't cost you anything but time.

If hardware does fail (and it will) as long as you have VCB / Agent based backups you can be back up and running on a new host in minutes.

If each of the current server roles are stand alone HA can't be a big concern as it is.
 
I think that may be a good place to start, and I will also put some sort of business case together and see where it goes.

Cheers boys!!! (and girls)
 
You must log in or register to reply here.
Back
Top Bottom