Comments on: Statspack Distractions

By: Analysing Statspack 13 « Oracle Scratchpad

Analysing Statspack 13 « Oracle Scratchpad — Mon, 07 Jan 2013 18:46:27 +0000

[...] final thought – it’s easy to fix the wrong thing when you’re looking at statspack reports. Although we can see some SQL that is [...]

By: Corretta informazione « Oracle and other

Corretta informazione « Oracle and other — Fri, 27 Aug 2010 08:46:00 +0000

[...] parole, ad un articolo di Don che fra l’altro riporta ampi tratti di un’altro articolo (http://jonathanlewis.wordpress.com/statspack-distractions/) fra l’altro, cosa che mi lascia basito, senza inserire un link alla fonte (sembrerebbe che [...]

By: Jonathan Lewis

Jonathan Lewis — Thu, 10 May 2007 15:35:29 +0000

John: a “hint” has always been a command that Oracle should obey.

However: hints are badly documented and very few people know how any of them really work; there are some real bugs that cause them to be lost; and Oracle has changed the syntax (without documentation) for some hints in 10g.

Bottom line – if you know what you are doing, hints will always be obeyed unless you happen to hit a bug.

By: John

John — Thu, 10 May 2007 10:56:33 +0000

You mention using hints.

Is it safe, in these days of 10g, to say that a hint is, in reality, a comand that Oracle will obey; or, is there still a chance that it will be ignored?

By: Jonathan Lewis

Jonathan Lewis — Thu, 21 Dec 2006 19:35:43 +0000

Prem, as an example of the concept: assume I have 10 tasks to complete in the night, but four of them have to run in the right order. It doesn’t really matter if I can make the other six jobs run in a few minutes – the time to complete the batch is the time it takes to run the four jobs one after the other. That sequence of jobs is the critical path.

By: Prem

Prem — Thu, 21 Dec 2006 13:09:00 +0000

Jonathan,

What do you mean “Critical Path” here ? I did not get that.

Regards,
Prem

By: Jonathan Lewis

Jonathan Lewis — Fri, 15 Dec 2006 21:39:54 +0000

Kevin, I don’t know, I didn’t check – I assumed they meant 4 processor.

By: kevinclosson

kevinclosson — Fri, 15 Dec 2006 20:29:51 +0000

Just out of curiousity, was this a 4 processor system? Or 4 core system?

By: Guy

Guy — Fri, 15 Dec 2006 17:09:56 +0000

I had a similar situation where it was critical that the “overnight batch processing” completed in 1 hour. This was a new system being put in (replacing 4 SQL servers with a single Oracle server) but because of the operational requirements (Bank), if there was an error in the batch, it had to be restartable and STILL complete within in 1 hour. Therefore, it had to complete in 30 minutes. Gulp.

But another operational requirement was that the system had to support 100% volume growth without hardware changes. So my batch window shrinks to 15 minutes!

This is where you start going very pale… 2 million accounts to recalculate in 15 minutes.

The good news was, exactly as you found, to remove as much as possible from the “critical path” and reduce the amount of processing required (only update the records that need updating… only calculate interest charges on accounts where the balance has changed – otherwise use last nights calculation)

The result was, with careful SQL analysis and profiling (for which Jonathan, Tom, Conor and Cary can take credit as I must have read every article written by them all!) we were able to get the critical path processing to around 8 minutes even on the busiest days without compromising data integrity, operational requirements or having to use any undocumented tricks.