Backup with vCentre and vSphere

[olouey001]

Hi,

I'm in the progress of planning our vSphere deployment to SAN. We've been using 3 * ESX Ent. 3.5 with Virtual Centre for close to 1.5yrs with local attached and now were moving to SAN storage. Were planning to upgrade our 4 ESX Ent 3.4 to vSphere 4.0 and then a run it with our EMC CX3-80 SAN with about 13TBs of storage with about 30VMs.

For our SAN sizing what I've estimated is the 30VMs will require about 9TBs of storage which we'll be running 300GB FC RAID5 disks and a remaining 4TB we'll use 1.0TB SATA RAID10 for the VM snapshot with VCB and scheduling the snapshot and then using our Tape Library attached to the SAN to backup to tape.

Can anyone give me some advice on this setup? I know for the backup there's also other alternatives such as using the EMC SAN snapshot but due to costs at this stage it not feasable at this time.

Many thanks

[AndreTheGiant]

VCB could be fine to do full image backup.

Be only sure to have VCB proxy connected to your FC SAN.

Cause you VM seem to be very huge (9TB for 30 VM) I suggest to split your backup in several jobs, so you will need a stage area smaller.

Remember also that VCB is only a framework. To make a backup to tape you will need also another program.

Andre

[olouey001]

Andre

Thanks for your reply. Yes we will be using VCB with our standard backup product Netbackup.

Additionally I'd like to ask for the vCentre and VCB because we will have three sites. Our Key site will have 4 ESX and 2 other smaller sites connected via WAN will have 3 -4 ESX each with SAN. Will there be any issue for the vCentre working with these smaller sites or is it better to have their own vCentre.

And what I was thinking was to use VCB at these sites as well but for each of these sites the proxy will be located at the site with FC tape drive attached to the local SAN.

[kooltechies]

vCenter should work perfectly fine with those distant sites , as you wrote these sites are smaller so there will not be much traffic going on between the VC database and the ESX servers. But you should make sure that the WAN link doesn't have any network latency.

Thanks,

Samir

P.S : If you think that the answer is helpful please consider rewarding points.

[olouey001]

Additonally I'd like to ask one thing any best practice on configuring RAID5 on the CX3-80 how many disks should I have in a RAID group? I'm planning to use just RAID5 for the VMFS but I'm not sure on the best number of disks to use in the RAID group and the size of the VMFS.

Any advice on metaLUNs using the EMC CX3-80 for best performance across multiple DAE's for VMware?

Thanks

[whynotq]

For best perfromance with R5 you are looking at 41 for capacity 81, yes meta LUNs across multiple enclosures will potential backend througput increase if your servers can actually use it. The IO foot print needs to be looked at, if you were to setup a 4+1 on 300GB 10K disks you've got a "rule of thumb" 640 IOPs to these disks so by doubling the size in a Meta you could get 1280 IOPs, but only if the VMs are actually pushing that kind of data.

I'd suggest gathering some workload stats over the coming week or 2 to see what you need, bare in mind that the 4 + 1 r5 is going to yield ~1040 GB so the Meta will be 2Tb, do you need this kind of individual size? if you decide to go with the meta i'd look at creating 2 LUNs in each RG and then extend the first one in the first RG into the first in the second RG, then extend the Second in the second RG using the second in the first RG. this specific route will give better IO distribution as Meta 1 will start in RG 1 and Meta 2 in RG 2, providing these are in seperate enclosures on seperate buses you'll be able to maximise the clariion backend.

You may want to consider PowerPath/VE if bandwidth is really critical but that, of course, will require "enterprise plus" but would give true multi-pathing.

[olouey001]

Whynotq,

Many thanks for the detailed reply. Actually another thing is are there any drawbacks from using metaLUNs that I will have on the overall?

Additionally how did you calculate the disk IO of 640 IOPs from 4+1 300GB RAID5 10K?

Cheers

[whynotq]

the drawbacks are negligable if you manage the VM distribution carefully and get the LUN layout acurate. the advantages, should the environment need it, is added resilience and extra IOPs.

the calculation is standard EMC:

5.4K drive ~ 80 IOPs

7.2K drive ~ 100 IOPs

10K drive ~ 140 IOPs

15K drive ~ 180 IOPs

although in practice and for burst IO we often see anything up to 3 times these figures but for sustained workloads the above is an accepted industry standard. so in the 4 + 1 R5 we multiply the IOPs per spindle by the number of data spindles 140 x 4 = 640 IOPs.

[DaveConverySHI]

For best results with NetBackup, the VCB proxy should also be a NBU Media server. Just install VCB and VMware Converter Standalone ont he VCB proxy and then NBU. You should have a local proxy for each site. You may find that you need to use two or three VCB proxies to make your backup window with such a large amount of data. NBU 7 will interact with the new vStorage APIs a little better and you will be able to have a de-duped backup with full vm and file level restores.

Check out my proven practice guide on VI:OPS for VCB -> http://viops.vmware.com/home/docs/DOC-1392

Dave Convery

VMware vExpert 2009

http://www.dailyhypervisor.com

Careful. We don't want to learn from this.

Bill Watterson, "Calvin and Hobbes"