Use fault tolerance to recover from disaster without downtime.. possible?

[comr]

Hi, I have two identical datacenters in diferent locations in my company and I'm planning to make them reduntant and high available, without downtime in case of failure. I was considering to buy a SRM license but I discovered that In case of failure you have to manually start a recovery plan in the recovery datacenter, and until this happens all the vms are down. So I'm thinking if it's possible to configure 2 groups of datastores in the same virtual datacenter, each group in diferent physical locations, enable FT for all vms, then store the vms in the first datastore group and store the secondary instances on the second datastore group, so in case of failure of the primary physical datacenter FT stars the secondary instances that are in the secondary physical datacenter. Is this viable?

[comr]

While researching, I discovered that vmware guys already thought about it. The solution called stretched clusters does exactly what I was planning to do, extend HA and FT funcionalities across two physical datacenters and manage them as a single logical datacenter.

[comr]

I've found that to use the stretched cluster solution I need to have a streched storage solution (which equallogic don't support) or use vsan stretched storage (which needs buying licenses, yet impossible for me). So let's get back to my initial question, is it viable?

[Nick_Andreev]

FT needs a dedicated 10Gb low latency link. Besides, FT generates a lot of synchronization traffic and creating FT for every VM will flood your network. This will simply not work.

There are products which let you implement active/active datacenters, such as VPLEX, but they are very expensive. Your best bet is to implement a DR solution using such products as SRM or Zerto.

[comr]

In each physical datacenter I have 3 servers with four 10Gb fiber channel nics, currently all my san traffic, including vmotion traffic don't even get close to 10gb. So I have 30gb of stanby network. What I was intending to do was dedicate two of the nics of each server for the normal san traffic and the other two for the ft traffic, so 20gb for ft traffic in each server in a separate physical network. Isn't it suficient?

[PemoK]

Hi  there,

I don't recommend you to use FT in this version. First of all you are limited for what, like 4 FT VMs per host MAX - and what is more important your performance will be 10x-100x worse than without FT. You can read this post or just google for performance problems around FT.

Fault Tolerance slow network performance

I think the feature isn't just ready for production YET (perhaps in 6.1 or 6.5)

greets

[comr]

Ok. So I think the best solution would be a stretched cluster, but I have Equallogic storages in my enviroment and from my research I think that only Compellent supports stretched cluster, is this true?

[Nick_Andreev · Accepted solution]

Yes, from Dell only the Compellent storage array is officially supported. You can proceed to VMware Hardware Compatibility List, select Storage/SAN category and in Array Test Configuration select iSCSI or FC Metro Cluster Storage to see the supported arrays.

However, even if you had a compatible vMSC array there would be a few other prerequisites. Metro Storage Clusters rely on synchronous array replication to prevent outage in case of an array failure. That obviously means that redundant low latency dark fibre would be required between the sites. Which adds up to the cost a bit.