Replication and failover on a T1 link...

Good afternoon, all!

I'm posting this afternoon to kick around a replication and failover scenario we're considering. The bare facts are thus:

Two DSS v5.3513 servers with both iSCSI and NAS volumes; one is at our main datacenter, the second will be installed at a branch office for disaster recovery. Replications are configured for both synchronous and asynchronous replications. The link between the two locations is a T1, which carries all branch traffic. We are Citrix oriented, so the actual traffic is generally screen and keyclick traffic.

The DSS servers house virtual servers running in Virtual Iron. We are adding XenServer into this mix soon; those Xen virtual machines will also run from the DSS servers.

First approximation: we set up an automatic iSCSI failover between the two nodes, using real IP addresses so router can move traffic with no additional overhead. Then we can test using manual failover procedures.

Second approximation: we set up a manual failover by "pulling the plug" on the main server we then promote the recovery server to production by stopping reverse replications and clearing metadata. We then set up iSCSI targets using the same configurations as the original production server and the Virtual Iron server. After testing, we reconfigure the recovery server back to replication mode, bring it back to our main center and resynchronize at wire speed rather than T1. Then reinstall it at the DR site and retest in another year.

Comments on what might not work or better ideas are welcome!

Thanks!

Gregg

That is very similar to the setup we are creating right now. We are lucky enough to have a 50mb/100mb pipe between sites, but we are also using Virtual Iron. Any particular reason you are throwing Xen into the mix? Is it because of the Oracle buyout?

Gregg,

With bandwidth costs the way they are today, I'm sure you can do better than a T1. If you get bandwidth from the same Tier 1 provider, you generally can pass traffic across their backbone at full speed.
I'd talk to Cogent or Internap and see what they can do.
The problem here how much data do you anticipate changing in a single second because that will surely lead to issues with your remote volume keeping consistent and therefore usable in the event of a problem.