Max number of CDP vm's

[william.bishop]

Curious question. What is the max number of vm's anyone has seen in a cdp layout? My understanding has always been its a specialty function for crit 1 servers, and I've never seen more than maybe 15 in use before. I ask because it was suggested we put 300+ vm's so that we could technically failover an entire datacenter....but I can't see it working to be honest. To me logically it would be replication for 95% of them, and 5% would be cdp. But I'm very open to being wrong on this one.

Thanks

[Gostev]

QA has recently tested 2K VMs with 5GB/s peak I/O load.

[william.bishop]

In a production environment?

[Gostev]

It's was a QA performance & load testing lab.

Apologies for a mix up:
The test I'm referring to above was for V12a that is still to be shipped.

The currently shipping version V12 was tested with 500 VMs x 4 disks each = 2000 disks total.

[william.bishop]

Maybe we can make it work....metro, so its close enough....Any tips to make it run better than it is currently is? It struggles to keep up, bandwidth is there, storage is there, hardware is all there. I'm thinking more cdp proxies, maybe specify source and target proxy.

[Gostev]

It's not clear what behavior exactly do you mean by "struggles to keep up". If there are not enough proxies, then proxy CPU would be the bottleneck. Is that what you're seeing?

[william.bishop]

I believe thats what I'm seeing, I'm just walking into this, I'm grabbing logs looking at the clusters. I only see 3 proxy's and they're sharing duty with regular proxies, that's where I plan to start.

[william.bishop]

I haven't worked with veeam software in about 9 years, I've heard about changes here and there, but was in a non-veeam shop. Lot of changes, I'm getting ambiguous responses, this group was told CDP was the future, by reps not engineers. And that very well may be the case, but not in this configuration.

[Gostev]

Indeed. For most customers, to protect the entire estate with CDP is similar to bringing machete to a kitchen... an overkill. Unless it is some special business, a big part of all VMs got to be fine with just daily backups. And may be only 10% of VMs (most critical) truly require CDP for near-zero RPO and RTO.

[william.bishop]

Thanks Gostev! I'm going to do my utmost to make it perform the magic that was portrayed, I had just never heard of this as a normal function, and was curious. It sounds like you guys are going in that direction, if you are testing 2k vm's on it though. Maybe replication will be superseded one day? Have a good weekend --wwb

[Gostev]

No, it will always been complimentary, even just because of long-term retention requirements alone... many companies do at least 3 years retention due to being a subject to various regulations requiring this.

[william.bishop]

Would multiple source and target cdp proxies be better than the rather oversized single cdp source? Layout is 7 rather large hosts, target cluster is an identical 7 host cluster. Metro multiple 10g between them. I feel like having the write filter be on the same hosts on both ends, ight help.

[Gostev]

Filters will be installed on all 14 hosts regardless, this is the requirement.

As for CDP proxies, it depends if they are physical or virtual, as well as redundancy considerations. If you do physical proxies connected directly into the storage fabric (as to not put additional load on your virtual infrastructure, impacting your consolidation ratio) then imho fewer oversized proxies is the way to go (as to reduce hardware footprint).

[william.bishop]

Its a vxrail on both sides, so not really an option to not do vm's as proxies. And my hope was having the filters on each host having a proxy on each to talk too...is it smart enough to use a local resource rather than filter writing out to a proxy on another host if there is one local?