Re: Large SMP Server -> Oracle 9i RAC on Linux

From: Daniel Morgan <damorgan_at_x.washington.edu>
Date: Wed, 03 Mar 2004 15:06:15 -0800
Message-ID: <1078355144.873280@yasure>

ctcgag_at_hotmail.com wrote:

> Daniel Morgan <damorgan_at_x.washington.edu> wrote:
> 

>>raj wrote:
>>
>>
>>>Anyone migrated Large (500GB+) Production Databases System from SMP
>>>Server to Linux and Oracle 9i RAC.
>>>
>>>What were the systems, what are the limitations encountered?
>>
>>Are you moving to RAC for reasons of scalability, fail-over, or both.
>>There are important distinctions.
>>
>>I've done it and the one thing I can tell you is that RAC has what
>>I call the 'spotlight' effect: It is extremely good at pointing out
>>very poorly written code. And often with the result that performance
>>goes down rather than up.
>>
>>I would be happy to answer specific questions but my generic advice
>>would be to have a very senior developer or DBA, someone that truly
>>understands how PL/SQL works look at the code and analyze it from the
>>standpoint of how blocks are read and might require transfer from node
>>to node. Last month I watched a RAC implementation take an app that
>>was doing 7000 inserts per second and drag it down to 1000 per second
>>because the underlying architecture was bad. A rewrite easily took it
>>to the stratosphere: But it did require a rewrite.

> 
> 
> Would those same changes have made the old SMP implementation go to the
> stratosphere also?  In other words, was the code bad only from a RAC
> perspective, or was it bad in general?
> 
> Thanks,
> 
> Xho

Yes and yes. Look at it this way. If I take a 4 CPU machine and do a build-up to 16 CPUs I don't have cache-fusion, I don't have to share blocks between nodes, but I do get to write a really really large check to my hardware vendor. Oracle gets paid for 12 more CPUs

If I take my 4 CPU machine and add 3 more to make an equivalent RAC cluster I Oracle still gets paid for 12 more CPUs and I pay very little for hardware but I do need to share blocks across the interconnect.

So the same issue would have limited scalability and performance on the SMP platform but without the block sharing across the interconnect issue. So yes ... bad code is bad code and hurts. But bad code with RAC is poison.

-- 
Daniel Morgan
http://www.outreach.washington.edu/ext/certificates/oad/oad_crs.asp
http://www.outreach.washington.edu/ext/certificates/aoa/aoa_crs.asp
damorgan_at_x.washington.edu
(replace 'x' with a 'u' to reply)

Received on Wed Mar 03 2004 - 17:06:15 CST