Distributed Backup? input needed please.

[steait]

Input needed for an idea for a customer.

Our goal is to create a very fast backup system from which we can run the entire production environment (instant restore) in case we want to.

We have a SAN with FC and iSCSI connectivity - a handful of ESX hosts which accesses datastores via FC.
Ok, now in each blade we'll put a local SSD disk of sufficient capacity and create a local datastore and place a Windows server with Veeam (proxy/transport/backup repo) - this VM will be tied to the host because it's on local storage. It will have access to datastores on SAN via iSCSI.

The idea is the create one Scale-out backup repo which will consist of the local backup repo of each VM in order to distribute the backup data evenly among the available datastores.

We'll enforce only backing up via direct storage access, we'll enable proxy affinity on the datastores to ensure the proxy/datastore is the same VM whenever possible.

Good idea? bad idea?

If good idea - what compression/dudup/scale-out settings are best (enterprise SSD class 1.8GB/s read, 1.0GB/s write and 250k/85k IOPS) - of course plenty CPU because each proxy can consume whatever the blade has as backups are run during low activity hours...

Input welcome..

[veremin]

Not the best practice actually.

A host goes down - you automatically loose access to backup data, and the time it takes to re-create a host, attach local datastore to it, spin up new VM, assign repository role to it and re-scan backup data, might be considerable.

Thanks!

[steait]

Ok fair point - we'd need at least 30 minutes to reinstall that host. Alternatively we could just move the disk to another host and get immediate access to the data (albeit we'd need hands-on).

Maybe I should include we'll keep a backup-copy around 50 miles away and we got a direct 10gbit connection - its just slow spinning disks.

Any other ideas?

[Andreas Neufert]

I think what you want to achive is more like a proactive restore situation.

What about creating backups somewhere outside of production hosts and then us VM replication (with these backups as source) to create VM replicas on the local host discs?

[steait]

Yeah not a bad idea, its what we have today more or less - it's just that it's spinning disks and the unit consumes a lot of power... I ran a few tests of my plan - unfortunately the blades I tested on don't have any write cache on the SAS controllers so the speed is quite slow. We're getting around 2GB/sec processing rate and 1GB/sec transfer rate - now nearly two hours into a 45TB full backup and about halfway... If there was cache I'd say the speed would double or triple at least - I guess best to increase compression and make the CPUs work harder. Anyway it is what it is.