Everybody Lies: Backup and Protected Computer Resources

Spent a few hours the other evening talking to a someone who in our industry about various backup vendors, products, and issues (Yes – you just got insight into what I do for fun. :)) This person works at a backup vendor and is pretty knowledgeable. Had a great conversation about the load that backup products put on a protected system in terms of the backup agent “footprint” – the CPU, memory, and I/O resource associated with backup.

We laughed about claims from vendors that agentless backup protection (and note, Unitrends does both agentless and agent) has a lower footprint. When you get really techie about it, and we were a little guilty of that, you start talking about protocols such as backup-specific and SMB/CIFS and how “chatty” the general-purpose protocols tend to be. But that wasn’t what surprised me.

We began talking about the benefits and disadvantages of doing VMware and Hyper-V backup at the GOS (Guest Operating System, or in the virtual machine itself) versus at the HOS (Host Operating System.) At one point the person noted two backup vendors (neither was Unitrends), and stated that one backup vendor who did backup in the GOS was superior because they only took “a few percent” of the CPU – and that the vendor who did the backup in the HOS could spike up to in the tens of percentage points.

Now – we’re all prisoners of our own reality – and this wasn’t the first time I had heard this story from a Windows-only backup vendor. So I asked questions like how many VMs were on the system, what the utilization of the system was, and the like. Well – those questions were difficult for the person to answer – because of their point of view.

I tell this story to illustrate a point. Companies like VMware have gone to a lot of trouble to optimize backup at the HOS level (note: Microsoft has gone to less trouble, so that backup vendors have to do a bit more, but Microsoft’s VSS backup architecture is at its heart a pretty innovative mechanism – and with things like primary NTFS deduplication coming in Windows 8 as well as replica handling it’s getting even more interesting.) So if you imagine you have 10 VMs, and you’re using backup in each virtual machine sucking away 5% of the resources of the system 24x7x52, you just put a pretty big tax on the server. Conversely, if you’re backing up at the HOS level, you’re using the inherently more efficient algorithms developed by the vendors within the virtual environment – so if you’re backing up once a day for an hour and taking 50% of the CPU (for example), you just gave a whole lot of CPU cycles over a 24-hour day back to the server or virtualization administrator.

Backing up at the HOS level isn’t always automatically better. In fact, I think that backup of virtualization environments works best when users have a choice as to whether to backup at the HOS or GOS level. But don’t be fooled by someone explaining “a few percent” to you without understanding the true impact on your physical server utilization.