Why does Veeam do serial VM backups instead of parallel?

[OREOSpeedwagon]

Just curious. I'm using Symantec PureDisk/NetBackup now for VM backups. It definitely has its bugs, but one thing I like is that it will fire off a configurable amount of parallel backups at once. So, I have a pretty large server (dual quad core, 32GB of ram, 8 NIC ports) connected to an Equallogic (4 active 1GB ports). When it comes time to do backups, the Symantec server seems to take full advantage of all the resources, maxing out all of the NIC connections and really hitting the RAM, and it will run a dozen or so VM backups at once.

I like this because we have about a hundred VMs and some of them are pretty large (1TB plus) and it takes a while for them to backup. If they were to be backed up in a serial fashion, one large VM create a pretty decent queue for the rest to wait in, pushing our backups into the next morning. As is, all of our jobs finish within our window, short of maybe one of the really large VMs which may bleed over a bit.

So, parallel backups just appear to be a better model from my experience. What's Veeam's take on this? Do you plan on sticking w/ your serial model? Why?

[Gostev]

We found that typically, single job is able to saturate the backup server. It is always either the storage connection saturated (for iSCSI storage), or backup server CPU saturated (for FC storage). For example, with (cheap) FC4 SAN in our own lab, quad core CPU of backup server stays at 100% with a single job.

Thus, we did not see really big need for this feature, and because there are hundreds of other features we need to prioritize, we did give this a high priority. Of course, we do understand that there may be some scenarios when the storage connection is not as ideally tuned as in the above-mentioned post, or with older storage devices in which case multiple processing threads may indeed help somewhat. So we may be adding an option of parallel processing in future. There are no technical limitations in our architecture that require us to stick with the serial processing.

Thanks!