Mulit-Site/Replication

[crucialx]

We are looking into the HP P4500 SAN, and trying to figure out the different replication scenarios supported.


We would have our main storage cluster at site A, and would like selected volumes to be replicated to site B and have XenServer to automatically failover those volumes replicated volumes to site B if site A were to fail.

How can this done with the HP P4500? Ideally we would not like both sites to have an identical SAN as we will only be replicating a small number of volumes to site B.

[Steven Clementi]

So...

Your Site A configuration can be slightly larger (or more) than your SIte B configuration. No big deal there.

Lefthand Replication consists of a Remote Copy Schedule... and well... remote copy's of data.

The shortest period of time a remote copy can take place is 30 minutes... so you can never have "real time" replication (synchronous).

Scenario...

Site A has 4 P4500 nodes.

If you only need to replicate 1/4th of the data, you would only need (theoretically) 1 P4500 node (of the same size) at Site B.

Getting the sizing of the 2 Storage Systems will be an important part of figuring this all out... as well as understanding what your rate of change for your data will be... and your schedule for remote copy's. The longer the interval... the more data that needs to be copied to the "off" site.


Steven

[crucialx]

Hi Steven,

Thanks for the detailed reply!

Is there a way to do synchronous replication in the origin method I've mentioned?

Ideally we would like the above setup using synchronous replication so as to ensure no data is lost. Site B will be within the 5ms latency required, and will have sufficient bandwidth for the replication (1G/10G).

Thanks!
Aaron

[Mark...]

Hi
To do this synchronously with P4000 means you would need to go to a multi-site config.
My understanding is that the max latency for this is 2ms (as stated by HP) not 5ms as you mentioned.
However to achieve a multi-site SAN you need to have the same amount of nodes at both sites.
As Stephen stated, scheduled remote IP copies can not be less than 30mins apart but you could run two schedules on the same volume.
Mark...

[crucialx]

Hi Mark,

At this point in time we don't have a Site B, we only have a Site A. Down the track we would like to add Site B, however purchasing the same capacity as site A would be cost prohibitive.

When we do setup Site B, we will ensure its within the recommended 2ms.

Do we need to setup a separate cluster or management group with each side of the multi-site cluster identical? I'm guessing we will have to setup a new cluster for this specific purpose so as to keep the number of storage nodes required down at each site.

On another note, how easy is to migrate a volume from one management group to another? Or one cluster to another?

Thanks!
Aaron

[Mark...]

Hi
For multi-site all the nodes are in one cluster.
If you wish to migrate a volume from one cluster to another then BOTH clusters must be in the same management group.
Note that if you do migrate a volume from one cluster to another that the VIP will change.
Mark...

[crucialx]

Hi Mark,

Thanks, that helps!

One more question, is it possible to change a cluster to a multi-site down the track after the SAN is up and running?

Regards,
Aaron

[Steven Clementi]

"One more question, is it possible to change a cluster to a multi-site down the track after the SAN is up and running?"

Since a multi-site cluster is simply multiple nodes split between sites... you can theoretically double up your storage and add in a set of nodes in another site. There will be some data re-distribution depending upon your configured network raid level(s).

Having a multi-site setup could effectively reduce the need to to "replicate" data to another site.. since your Network RAID level would potentially automatically copy data to the other site. (which is the reason for the stringant 2ms latency or less.)


Steven

[Mark...]

Hi,
The answer is yes you can convert to a single/campus SAN to a multi-site at a later date if you wish.
However, it is important that you get the cluster stripe order correct or it will not be as fault tolerant as you may think!
If you use the CMC wizard to create a multi-site SAN then it does it all for you, but to do that you would have to delete your existing management group which would mean all your volumes = data = pain!
So when you convert your single site to multi-site you need to make sure that the stripe order is correct.
For example let us take four nodes called 1,2,3 & 4 all on one site in that order from the cluster properties page in CMC and network raid-10 for all volumes. When you convert to multi-site you should make sure that 1 & 3 are on site -A and 2 & 4 are on site-B. Then you should be OK.
Mark...