Bridged Single IP Subnets for MC/Service Guard

[Krishnan Seshadri]

Hi
We are in the process of separating our datacenters, splitting HPUX servers between the two dc's and implementing MC/Service Guard clusters between the two buildings. We are also in the process of implementing a Layer3 CISCO network architecture migrating from a layer2 nortel scenario.

Recently we have been told by our datacomm folks as well the networking vendor that it is impossible for them to provide us the extended same IP subnet between the two datacenters (without compromising the sanity of the core network architecture), which is a underlying major requirement for MC/Service Guard. It seems kind of "unbelievable" to me that a major networking architecture doesn't have room to provide such a common H/A feature. We are also planning to implement MS stretch clusters for our W2K servers in the near future, which also is in jeopardy because of this development.

If any of you have implemented such networking architecture with MC/SG implemented (in extended campus or metro scenario) between two dc's, pl help me out.

Krishnan Seshadri

[Michael Steele_2]

This is the basic underline for disaster recovery sites using and failovers between data centers.

One common implementation is for two (* XP 1024 *) disk arrays using BCV's sync up over a SAN switch with 'long wave ports' (* I believe its called *). There are two nodes in either data center making four nodes altogether.

Since 9/11 all utility companies have been given a mandate by their regulatory agencies to implement similar schemes or face some kind of penalty.

Direct your search into SAN architecture as well as 'continuous access' discussions. Here's one link that may be helpful:

http://docs.hp.com/cgi-bin/fsearch/framedisplay?top=/hpux/onlinedocs/B7660-90010/B7660-90010_top.html&con=/hpux/onlinedocs/B7660-90010/00/00/32-con.html&toc=/hpux/onlinedocs/B7660-90010/00/00/32-toc.html&searchterms=copy%7cvolumes%7cbusiness&queryid=20030419-053542

[Rita C Workman]

You can go Metro Cluster using either CA technology (HP) or SRDF technology (EMC) to data replicate between your disk arrays. This type of technology is what is needed to accomplish what I 'think' you are attempting.

Now depending on distance you might need something more than a Metro cluster. A Continental Cluster would be required if you are beyond the distance limits of a Metro. Once you go to a CCluster you now have to address the fact that you must failover to another subnet...and they requires alot more hoops to jump through. But sounds like your more campus/metro based.

So the question is what software are you using to replicate your data between disk arrays !?!.

Just a thought,
Rita

[Krishnan Seshadri]

Thanks Michael & Rita

We are implementing SRDF to do the data replication and the distance (fibre run) between the two datacenters is just 27miles. Other than my question about the "networking", we have everything else ironed out. All along we were under the impression that the Single IP subnet between the two datacenters is a "no brainer". And now we hear that it is something complex to achieve due to "re-convergence" issues incase of network switch failures. Any way thanks for your reply and I will keep my eyes and ears open.

Regards
Krishnan