Wednesday, June 18, 2008

Block Size and Oracle Database Performance

Recently, on the Oracle OTN Forums, there has been a major debate over the
Oracle block size and database performance. Some top Oracle experts including Jonathan Lewis and Steve Karam have joined the debate. More details can be found here and also here on the issue of whether or not block size for Oracle determines whether or not performance can be improved or not.

In a nutshell, some of the discussion touched on potential bugs in Oracle database code with respect to the use of ASSM and block sizes.

I may or may not create some actual benchmark tests to see how my results differ from the financial services customer I worked a few years ago to resolve the poor performance in the past. I saw major improvement by changing the block size from 8k to 16k. Due to NDA and non-disclosure laws, I was unable to post the actual statistics. Also, one item from the above block size and Oracle performance debate that has been interesting to note is we have two groups of thought at odds with each other: one such as Jonathan Lewis and Tom Kyte advocate extensive test cases and posting results to argue a point on Oracle and the second group represented by Steve Karam and Don Burleson advocate real world cases to prove how Oracle database really works in the real world.

It really depends on how much time I have.

Personally, both camps of thought have some interesting points to consider. I argue this because testing is important and actual results from real world production systems are critical as well. However actual benchmark testing from at the bare minimum a subset of a live production system appears to be more accurate than a simple test case.

No comments: