Oracle and the SAN

[xmario2013]

Hi All:
We are thinking about putting Oracle data file onto a SAN which the storage will be separated from the server itself, is it better to setup like that anyway ?

Also we need to setup replication and fail over between two sites , the SAN vendor can offer replication software to snapshot or just hot-replicate the data, I am assuming we need a second Oracle server to handle the replication and failed over data to redirect user and ensure up time.

so we need to install oracle on the second server and create/restore the same database name ?

(ie. we are using NT server and we dont have OS-based Cluster)

does anybody have experience in doing this ?
how we should go about and do that ?

Thanks
XM

 

[markrem]

I have had this a few times. Putting on the SAN is not necessarily to improve performance, but is can if you pay the big bucks. My last price quote from a vendor was around $10 Million per device, and the disk array had nearly 100 CPUs and 100 GB of RAM, in addition all of the CPUs and RAM in our database server. The SAN box works like an enormous SGA. The entire database, all 10 TB, was cached in RAM. That's some fast stuff.

Also think capacity. You might have 200 disks at 200 GB each in you SAN box. Even if you use RAID 10, that's still 20 TB of storage space in a single box. How many disks can a Sun v440 hold?

Regarding your question of whether the databases will have the same name or different name purely depends on which type of replication you setup. In EMC SRDF, they would have the same name and be the same database, just on different SAN boxes with different front end servers using their own IP addresses. In other replication technologies, the databases might have distinct names. It all depends on which way you go.

In terms of replication, I've used one of the vendor provided replication techniques and walked away from others. Oracle would respond best after a failover if you had full-hot-block syncing, but then your servers would need to be within 60 KM of each other. Not good earthquake protection. There is semi-sync replication at the hardware level, which replicats batches of disk blocks at a time. The worst case scenario is you lose the last N blocks, which hoses your control files and you therefore lose the contents of your on-line redo logs (so keep them small.) Async replication at the hardware level is not supported by ORacle and I can personally tell you it is impossible to get a complete recovery if the thing crashes. You control files are hosed, and you have to recreate them and the on-line redo logs every time. Ugly, so stick with full hot sync if you can.

In terms of other types of replication, checkout DataMirror and Quest. Both use their own log scrapers, similar to Oracle Data Guard, except DataMirror and Quest make it a ton easier to bounce back and forth between the production and standby instances (Data Guard is really ugly on this.)

Oracle Advanced Replication is another route you can take. It's slowest of all options, but once you get it running can do the job. It's being phased out in favor of Oracle Streams, which I think was announced in Oracle 9.2, but Advanced Replication is still here in 10.1.

-Mark

9i OCP