Replication Target Host selection

[LarsWulf]

Hi everybody,

I am currently evaluating Veeam as Replication solution for a DRP project.
Everything is fine so far but I notice one suboptimal behavior. All test jobs are configured with the vSphere cluster as target. But always just one host is used. So I am quite quickly stuck at waiting for ESX Target availability due to nbd limits per ESX host.

Is there a way to separate the jobs to different hosts without configuring different jobs with dedicated hosts as target.

Regards
Lars

[foggy]

Hi Lars, why don't you use hotadd instead of NBD to populate VM data on target?

[LarsWulf]

Normally I like to keep the load of backup and restore at dedicated machines used this. So that is why we are using physical servers.

So to be honest I didn`t really thought about a virtual proxy. But good hint.

Are there no limitations or bugs (not mentioned in kb1054) when you are using HotAdd ?

[foggy]

All the limitations are described in the mentioned KB.

[LarsWulf]

I will give it a try.

But just for information: why do I have the described behaviour with nbd ?
How is Veeam choosing which host is used out of a cluster ?

[foggy]

It selects the host randomly. In 9.5, however, some load balancing logic will be introduced.

[mbrownwfp]

I will preface this question by saying that we run vSphere Standard on the majority of our hosts, so DRS is not enabled.

I've noticed that if I select a cluster as the destination for a replication job containing multiple VMs, all of the replicas get created on the same host in the target cluster. Is this due to lack of DRS at the VMware level or is Veeam making this decision independently?

If this is native Veeam behaviour I am struggling to see the benefit of allowing a cluster to be chosen as the target for a replication job if there is no distribution of VMs across said cluster.

If distribution is dependent on DRS it would be helpful for Veeam GUI to indicate there will be no distribution within the cluster due to lack of DRS, or simply not allow a cluster to be chosen as a target if DRS is not available (similar to how VMware wont allow a cluster to be chosen as a vMotion target if DRS is not available).

Hoping someone can provide some more detail on this behaviour so I can plan my replication jobs accordingly.

Thanks,
Matt

[foggy]

Hi Matt, the behavior is expected with the current version, but will change in v9.5.

[mbrownwfp]

Thank you foggy.

You mentioned the target host is selected at random. I appears that the host selection happens at the job level (rather than for each VM in the job) as each of the 13 replicas in my job ended up on the same host. Random selection even on a per-VM basis within a job would help as this would provide some level of distribution for the replicas.

Anyway, I look forward to this and other new features coming in v9.5.

Matt

[foggy]

I appears that the host selection happens at the job level (rather than for each VM in the job) as each of the 13 replicas in my job ended up on the same host.

Right, this is how it works currently.