Re: What blocksize?

In article <HQyu3.13553$vu2.3645_at_news.rdc1.tx.home.com>, Bert Scalzo bscalzo_at_home.com says...
> That said, pick 4096 for OLTP.
> If you had been a DW, then 8192 would be more appropriate. The old
> default of 2048 is a default from way back and is no longer a good choice

I find it hard to decide which blocksize is best. The books and documentation I have read is very vague about this.

Why is 4096 better than 2048 if you have a system which most of the time only access few rows and don't use much full table scans? And how about the db_file_multiblock_read_count parameter? With the defaults in the starter database with blocksize=2048 and db_file_multiblock_read_count=8 means that 16kb data is fetched for each i/o.

Wouldn't increasing the db_file_multiblock_read_count give the same benefits as having a larger blocksize?

Another questions: Should blocksize (or maybe more correctly blocksize * db_file_multiblock_read_count) match the physical cluster size on the disk?

Magnus Received on Mon Aug 23 1999 - 14:48:21 CDT