High availability with two FAS 2552

[IMPI]

Hi,

first of all i need a definition : what is a node ?

A fas 2552 with two controllers and 2x12 disks is it one or two nodes ?? I guess two, am i right ?

This said :

we have two rooms separated from about 30m (then could be connected with a local and fast network) and in each we would like to have one fas2552 and two ESX.

This should work as PCA and the loss of one room should not avoid the other to be fully functionnal (ok for esx, here my question is only about FAS)

Are we able to connect both 2552 using HA or HA is reserved only between both controller of the same FAS ? How ? Is there more software to buy ?

I was told about using snapmirror between first room and second room, but for me this will only gives me PRA not PCA, as both netapp fas will be seen as a separated, not as one ?

(There will also be a remote site for PRA but this will be handled by Veeam the no question about it.)

Thnaks for your help

[aborzenkov]

Node is controller. HA pair always has two controllers (nodes), in your case in the same chassis. To get transparent failover between two locations you need metro cluster, but metro cluster is not possible with FAS2xxx. Alternative is host-based mirroring (simply allocate storage from both HA pairs to VM and mirror on VM level).

[IMPI]

Waouh ! such a quick reply, many thanks.

Can you please tell me more about your alternative solution ? I can figure it out

If i want a transparent failover between my two rooms, what are the minimum requiremets ?

[SeanHatfield]

To acheive that with the equipment you have, look at using VMware Fault Tolerance (FT) at the virtal machine level. 

[IMPI]

Ok thanks, i'll give it a look. I do not have bought anything yet, still under thinking about the solution !

but i read and reread first solution provided and i guess i understood something :

in fact, i'll configure two disks for the VM one on first FAS and the other on second FAS.

Then i'll have to do a kind of raid1 between theses disks (either using windows soft raid, either using an esx feature i do not know, and that you're talking about).

then i should forgot snapmirror

I guess a real PCA solution should be better...

[SeanHatfield · Accepted solution]

It depends on your objectives, and your budget.  If you want "continuous" availability, use FT in vsphere, or application level continuous availability solutions.  If you want synchronous replication between rooms, or between sites within a metro area, consider a metrocluster solution.  If you want DR to a remote site, use snapmirror and evaluate tools that orchestrate the recovery workflow at the DR site.

[IMPI]

Once again many thanks to both of you !

Yes i need to have VM working with no interruption if i loose one room. Servers will be sized to be able to handle all VM on only half of them.

This means FT for ESX, sure ! At this point this become a vmware thread 🙂

But i also need storage to be able to support the lost of one room.

I can see that this is not native with netapp 2500, i need either another software layer to "virtualize" both FAS into one failsafe storage or go to metrocluster with another FAS serie.

Just please tell me if my understanding is correct

And thanks for replying during a week-end 😉

[IMPI]

Sorry, i just see i did not updated this thread.

To achieve storage failover we went over a "strech metro cluster" solution, less expensive than a metrocluster solution, but limited to 100-150 m between rooms. This is ok for US.

This will be done using 2xFAS8020 controllers and 2xDS2246 shelves.

Now the configuration for that will be another pain 🙂

Thanks

[aborzenkov]

I tend to miss the fact that in ESX virtual machine definition can still reside on a single datastore only. So if this datastore is lost, you lose your virtual machines.

You probably can build virtual datastore based on DRBD or similar, but at this point there is not much of filer left, it becomes pure ESX topic.