Re: Block size issue

It is stated in the Oracle 7 Data Base Administrators Guide that larger is generally faster. Based on experience, I would agree.

I/O is the slowest thing Oracle does and larger Oracle blocks generally results in less I/O requests. Also if you look at the statistics generated by the utlestat script you can determine what percentage of your I/O is sequential. Ours runs around 60% sequential in a transaction driven system. Non-unique index use often results in sequential reads of the index blocks, the cost optomizer will do full table scans on small tables rather than read the index then read the data block when reading all the data appears to be cheaper, and batch jobs (including exports and sqlldr) benefit from larger block sizes.

The biggest cost to large blocks may well be the demand for Rollback segment extents that results. Bigger blocks increase the amount of rollback area required, but we have found the larger the database the better it is to have larger blocks. The reduction in Oracle overhead from each table can be significiant for larger tables.

Food for thought. We have 60G of data and over 1000 users though only around
270 at a time.

Mark -- No witty disclaimer Received on Thu Mar 21 1996 - 00:00:00 CET