VDP vs VDP Advanced - Backup Speed

[MC_jeffhoward00]

Hello -

Does anyone know if VDP Advanced is any faster than the 'regular' free version?  We finally finished our Production deployment of the 'free' version of VDP, and it on days with heavy data change (higher than average client processing for our application), it has a tough time finishing within the backup window.  For VDP appliances that have to backup VMs over the network (meaning the VDP appliance and VM are not on the same hypervisor), we regularly fail due to the backup window cut-off.

My thinking is, if we were to upgrade to Advanced (for the extra capacity), I don't think we could get 4TB of data through the appliance on a daily backups schedule (definitely no chance of using the 8TB version in our environment).  I think we'd have to create a staggered every-other day schedules (or something similar).  What we can't figure out is why it's slow - As far as we can tell, nothing is maxed out - CPU stays below 50%, and Disk I/O seems lower than other VMs on the same Hypervisor.  Possibly it's the complexity of the Disk I/O due to the nature of de-duplication?

Any suggestions for increasing speed on their version would be appreciated as well.

Thanks,

- Jeff

[Paul11]

I would try to give the appliance more memory. I'm using 16 GB of RAM and have good performance. I don't think there is a performance difference between VDP and VDPA. VDPA also defines more memory on the higher capacity versions. Using more appliances could also be a solution for your problem

Paul

[schepp]

MC_jeffhoward001 wrote:

For VDP appliances that have to backup VMs over the network (meaning the VDP appliance and VM are not on the same hypervisor), we regularly fail due to the backup window cut-off.

It has nothing to do with being on the same host. It's about the host where the VDP-appliance runs having access to the storage where the target VM is stored.

If it's not already configured in your environment, you could speed up the backup process by giving the host where the VDP appliance is running access to all datastores, so that the backup doesn't have to be transported over the network.

For example: I have a SAN shared across 8 ESXi hosts. VDP will snapshot the target VMs on all 8 hosts and mount the snapshot to the appliance and do a local copy without any network traffic.

                    Then there is one host in another location that has no access to the SAN. The VMs on this host are located on local storage and will be backup up  over the network using NBD.

The over-the-network backups are really slow, I get a top transfer rate of 20MB/s over a 1GB connection. Don't know why, maybe it's a QoS type limit from VMware to not totally block the ESXi hosts management network.

Regards

Tim

[MC_jeffhoward00]

Thanks Paul, I will definitely try that.  Makes sense that being able to store larger contiguous spans of data in memory would allow the de-dup engine to run more efficiently.  I'm logging historical job run-times and "new bytes" using the MCCLI.  I'll report the results back to this thread to give people some pragmatic evidence of gains from increasing memory.  As far as I can tell "new bytes" seems to be the most accurate indicator of how long a job will run.

[MC_jeffhoward00]

To clarify, in our environment it does matter which host it's on   If it's on the same host, then VDP is guaranteed to have direct access to the target VM storage whether the storage is direct-attached or SAN storage.  I think what you're saying is if all hosts are SAN-attached, then it doesn't matter what host VDP is on, as long as there's a valid storage path from the VDP host to the target VM host.

That is true for us in Production, but we still use direct-attached storage in pre-production.  Hence VDP either has to be on the same host, or it will fall-back to network based backups if it's attempting to backup over the network, which has proven to be impractical for the VM sizes we use (1-2TB of storage per VM).

Thanks,

- Jeff

[schepp]

Hi Jeff,

yes VDP's network backup is not the right choice for 1-2 TB VMs

Depending on the number and size of your preproduction servers, you could think about mounting a NFS share on the hosts and deploy a VDP appliance per host on that NFS datastore. This way the appliances would not use network backup.

As said it really depends on the size of your infrastructure if this is a practical workaround for you.

Tim