Because backup of a vSphere VM almost always involves taking a vSphere snapshot, this VMware blog post will be interesting for every backup administrator.
VMware has tested the performance impact of snapshots. Baseline performance is a VM without a snapshot. After that, performance testing is done with 1, 2 and more snapshots. Tested was default IO-tests and java application performance (SPECjbb). Tests included: vVOL, VMFS and vSAN.
- Impact on vVOL depends on the storage system, because snapshots are taken there.
- VMs on VMFS have a huge performance penalty even with one snapshot. To be more exact: the first snapshot has the greatest impact!
- vSAN does not suffer much from snapshot with sequential workload. To be honest I think this is interesting to know but has no meaning in reality.
- SPECjbb does not show worse performance at all.
- Let snapshots exist as short as possible.
- vVOL snapshots create less impact - but it depends on the storage array.
- Keep the snapshot chain as short as possible.
I recommend to read this post carefully (5 minutes). At least the graphics in the post are good to show to the guy who always forgets to delete his snapshots.