Scaling Oracle for developers

There are alot of books and articles on Oracle performance tuning, but this book filled in some gaps. I've done alot of Oracle Performance tuning, mostly at the application and database configuration level. Mostly providing the conduit/moderator between the developers and the dbas and sysadmins. I've never really investigated in detail one of the most crucial factors, disk configuration and this was the only book I could find that looked like cutting through all the layers (application, db, os, kernel and architecture) at a level I was comfortable with.

I first saw the author a few years ago at a Miracle A/S event, the most useful Oracle conference I've been too, as it was purely for Performance Tuning based.

This is an old book that was released in 1999, so some topics have been renamed/superseded but the concepts remain.

Part 1: Concepts and Architecture.

Chapter 1 provided a good discussion of Scaling Concepts within Oracle.

Good description of Architectures, CPU and I/O subsystems, including RAID configurations. This part was the main reason I bought the book and provided me with the level of detail I was after.

See asktom for more details of recommended raid configurations for the different types of oracle files, Raid 0+1 is generally good for all types (although you could skip it on redo logs), forget raid 5 if you want performance and have a reasonable amount of write txns. Parity calculations are too costly.

Part 2: Building Support Software

Benchmark Concepts and Design: was a good chapter in terms of overall approach. The tools could still be used, but it wouldn't be my choice. Particularly if you are interested in benchmarking web site performance backed by Oracle.

System/Database Monitoring:
Alot of this chapter is based on monitoring v$ tables, or Oracle Wait Events. 10g has a great console for performance monitoring, it is so much easier to find problem areas. You may use additional tracing in specific once these are identified. Too many pages on formatting measurement data in excel.

Disappointing to see that collection of stats was only from the database.

Benchmark results always need to be considered in relation to what is going on within the application, the database, the os (cpu, io and mem) and the network.

Collecting stats from these area's is easily done with unix skills and even easier done with 10g that reports on some os metrics like cpu utilization. Other oracle stats can point to os issues, like excessive io wait times points to poor tablespace or disk configuration.

As well as collecting stats you also need to ensure things are not failing, check log files in your appserver, db event logs to ensure that things are failing, since they always run faster.

See also http://www.hotsos.com for details of instance tuning and SQL performance tuning in relation to Logical IO.

Part 3: How Oracle works

Good if you don't know much about Oracle internals, although this is pretty widely documented from Oracle.

Part 4: How Unix Works

An Introduction to Unix architecture, nothing you couldn't get from other sources. Chapter 8 was a dud, the unix tools description was pathetic.

Part 5: Implementing Oracle

Scalable Transaction Processing using TP monitors is not as relevant nowdays since application servers are taking over that space.