Sunday, June 20, 2010

Exadata DBA

Well I guess I am like thosuands of DBAs out there, who are intrigued, excited, curious, and yearning to lay their hands on Exadata, which looks sure enough the future of OLTP and DW for the large and even medium size enterprises across the globe.

What should a typical DBA who is aware of core database, RAC, Data Guard, ASM should do to learn about Exadata without having it?

Exadata is kind of "different Oracle", for example it doesn't require indexing on table that much and so and so. So the DBAs need to re-define themselves to manage the Exadata?

Yes, storage concepts and the hardware concepts do matter when it comes to manage the Exadata, but how much knowledge in this regard is expected from the DBAs?

What are the responsibilities of an Exadata DBA?

I believe that some of above questions overlap, but would appreciate any answers in that regard.

And I got following responses from the experts:

An exdata DBA need experience in Database, Storage and OS.

A good understanding of SmartScan and Hybrid Columnar Compression. (ctrieb)

If you can answer the question "Why with Exadata do I probably not need [some or all of my] indexes?" you are on the right road. (Dan Morgan)

Learn all you can about direct path reads as they are critical to enable smart scan. Serial direct path reads are done often in 11gR2, probably because of Exadata influence.
Learn about parallel query and the mechanisms available to throttle it (queuing is now available in 11gR2).
ASM provides the redundancy on the disks, so make sure you are familiar with the options provided.
Knowing something about Infiniband would probably be a good idea as well since you'll have to figure out how to connect to external devices (tape drives for example).

Read all the Oracle published white papers (there are quite a few out there now).
There are a few blogs with some info: Kevin Closson has several posts on his blog (although he's kind of fallen off the Exadata wagon lately - but we all have real jobs), I have posted a few as well. There will be more information coming available as more people get access to them. Oracle seems to be selling them like hotcakes.

Throwing all the indexes out will probably not be the right approach for most mixed workloads, by the way. Although migrating to Exadata may provide an opportunity to get rid of a bunch of unneeded indexes. A lot of systems have a bunch of indexes that aren't needed to begin with. (Kerry Osborne)

No comments: