MetroCluster solution needed?

[Thomas J. Harrold]

I need to design an HA solution for an app that uses an Oracle database on the back end. I have two sites that are approximately 15 miles apart. Desired solution will allow application clients to run from either site. Should either site go down, failover should be as seamless as possible. Can an Oracle RAC cluster run over such a distance?

I'm concerned about the distance. I'm very familiar with clustering, but have never had to "extend" a cluster to a second site. Can anyone give me some limitations? I haven't found much other than fluffy marketing docs about MetroCluster. Any decent docs/whitepapers would also be appreciated.

Thanks in advance...

-tjh

[melvyn burnard]

Here is the Designing Disaster Tolerant High Availability Clusters manual URL:
http://docs.hp.com/en/B7660-90016/B7660-90016.pdf

It shoul dhelp you with your questions.

[Stephen Doud]

A Metrocluster configuration requires 2 storage arrays which continuously copy changes to writable disks to the remote duplicate - somewhat like LVM mirroring but under control of the array software.

Melvyn refered to the document - and I located a web-page in that document which describes cluster types and when you would go from an Extended Distance Cluster to a Metrocluster:
http://docs.hp.com/en/B7660-90016/ch01s03.html?btnNext=next%A0%BB

This section gives distance limitations.
-sd

[Jan van den Ende]

Thomas,

at the risk of being totally off target:

Oracle runs just fine, with simultanious access, either with site-preference or load balancing, on a two- or three-site VMS cluster, synchronously shadowed storage, with total distance (measured around the actual ring) of up to 1600 KM (guess you call that 1000 M).

The big question: How much of your app is Oracle itself, and how much is "OS-specific" code around it.
It the first is the great majority, it is a REAL GOOD option. Many such configs have already celebrated their uptime lustrum. :-)

If a lot of code is OS-specific, then the cost of migration could make my suggestion of lesser value. :-(

fwiw,

Proost.

Have one on me.

Jan

[Thomas J. Harrold]

Jan, thanks for the suggestion. I'm willing to consider all options. The app we are supporting did originate from VMS, so I'll never say never....

All: I am seeing some discrepancies in the HP documentation. Can FC switches talk up to 50km without going to DWDM, or is it only 10km? (I've seen both numbers...)

I believe our "ideal" solution would allow application clients operating at two locations (site A and B) to access the server running at site A. If site A went down, remaining clients would access application from site B. (with no manual intervention required).

Again, I know that conceptually this is possible, but at a 15mi distance, I'm not sure what performance will be like.

Anybody have any real world experience with extended cluster?

-tjh

[Jan van den Ende]

Thomas,

Real world experience?

Yes, but (of course) not exactly your params.

Our inter-site distance is only 7KM (round trip 15.5 KM),
And NO native-Oracle database, but Oracle-RDB and Oracle-dbms (and Progress, and Basis+, but those are regrettably not cluster-wide, just failover).

We have NO latency problems whatsoever.

My experience with native-Oracle dates back to VMS V6.x with ORA V7.y, and that used to also run smoothly cluster-wide.

Inhowfar that info may be extrapolated is beyond my knowledge.

hth,

Proost.

Have one on me.

Jan