Considerations of Multiple Replication Job with 15 minute RPOs

[joebranca]

We are doing a feasability anaylsis of replicating about 40-45 VMs (mostly application / fornt end) to a remote cluster every 15 minutes 24/7 (except for the VM backup windows).

Besides the actual data change deltas in those 15 minute periods (which should typically be minimal under normal conditions), what are the major considerations on meeting the RPO with this many VMWare snapshot creations / deletions?

If we setup either say 3 jobs of ~15 VM replciations running every 5 minutes, or 5 jobs of ~9 VMs every 3 minutes, has anyone seen impact on the environment when a certain number of vSphere snapshot operations are concurrently running, either on the storage IO, ESX host resource level, and/or performance on vcenter itself? Do vSphere snapshots queue up at a certain noticable point that could impact the ability for 15 minute intervals to be reached? Or does it completely depend on how the VMs are distributed among the source cluster's hosts and/or datastores?

[HannesK]

Hello,
assuming that you have enough IO network performance, the VMware snapshot creation / deletion is the main consideration, yes.

Bigger or smaller jobs are not really relevant. Enough proxy tasks (CPU cores) must be available.

Whether you put 10, 12 or 45 VMs in one job does not really matter. The scheduler will handle it.

Without registry keys tweaks, Veeam will handle the amount of snapshots at a general best practice level (4 per datastore at the same time). I would start with the default settings.

To have some safety margin, I would start the jobs every 7-10min and see how it goes in your environment. I would also keep the number of restore points small to avoid long snapshot consolidation at the target.

VCenter should be no problem with 45 VMs.

Best regards,
Hannes

[joebranca]

This is very helpful, thank you for the information.

We have 10Gb networking across the sites and use NBD mode for every other Veeam operation.

One follow up question since I apologize I'm a little uncler on proxy tasks for replications. When setting up dedicated proxies on both sites for these jobs, the closer we match the number of CPU cores / concurrent tasks to the number of VMs' disks in a job that runs at any given time will minimize delays in replication processing for that particular job?

[foggy]

Hi Joe, that is correct.