Re: ASM for single-instance 11g db server?

> Moral of the story is that we as DBA's need to fully understand the
> entire "system" when dealing with performance. On many occasions I
> have encouraged those wanting to be a DBA to also learn the
> Sys,SAN,Net admin jobs at least sufficiently enough to know when they
> didn't configure stuff correctly. Especially in a RAC environment.
> [quote from me on more than a few environments "NO, do it the way I
> told you to do it and not the way you think it ought to be done -
> especially if you want it to actually work"]

Absolutely! Those extra factors are a lot more significant nowadays - assuming recent hardware and software - than the odd msec here and there between RAID levels.

Look, a couple of years ago we hit a wall with our then new CX380, upgraded from a CX340. It could do better, and we knew because we did measure the performance inside the SAN and it was there. But from Aix? No way Josay. After much head scratching and banging of walls, we figured out what was ticking.

Our DC is leased. The new SAN had been located in the 4th floor. The main Aix server was on the 14th. When we questioned the leasing company as to the why of such stupidity, we were told that the Brocade fabric connecting the whole lot "was fibre optic and nothing is faster than light, you know that!"

Yeah! Sure...

Apparently no one told these geniuses when we are dealing with 500MB/s and higher bandwidth, light travels only 60cm for every data bit. Or around 4.8 metres for each byte. And between each byte and the next there has to be a signal separator, which happens to introduce a discontinuance to an otherwise sequential process. And acks/naks (or their equivalent in Brocade parlance) become also a LOT slower!

Now call me strange, but I contend there is a *lot* more than 4.8 metres between a 4th floor and a 14th floor! Let's not forget that the fabric was not point to point either: it wasn't exactly just a single fibre cable stretched between the two floors.

Once we adjusted for all those distance losses, the reason for the "wall" we hit was perfectly clear.

And we proved the point later on with our DR site where the SAN is right next to the main Aix server and is hence capable of a reasonably higher non-sequential data rate, without any spurious interference. There we hit the theoretical limit of 800MB/s of the Brocade connectivity we had, without killing anything else.

Now tell me:
which part of the OCP gives ANYONE the preparation to address these type of problems?
Which screen in grid gives ANYONE the ability to detect this?

Any wonder why I consider those certifications and tools as just noise when selecting anyone or anything for a modern dba job? Received on Wed Apr 06 2011 - 07:24:00 CDT