Re: The rule of four for rbs segments -- HELP!!!!!

1. Don't post in HTML.
2. It's not a rule, more a rule of thumb
3. 1 per 4 concurrent transactions is general for average systems, but it usually degrades to around 1 per 8 for particularly busy systems

The issue is one of contention for the Rollback Segment Header block. Since all transactions have to inspect that block in order to find out where to place their undo, on a busy OLTP system with lots of concurrent transactions, there will be a queue forming for access to that block. Having more segments mean you have more segment header blocks, and accordingly transactions aren't now going to fight over a scarce resource.

In batch update systems, the number of segments is almost irrelevant -on the grounds that you don't tend to fire off simultaneous big, bulky updates (not if you've got any sense, that is). There is therefore little liklihood of contention. Hence fewer segments are needed. However, big, bulky transactions require big, bulky rollback segments -you don't want a transaction waiting whilst an under-sized segment has to acquire additional extents.

Regards
HJR ""Stefano Mingoli"" <smingoli_at_selfin.it> wrote in message news:000901c0a286$2e03c410$922a500a_at_smingoli... Hi there,
can somebody please explain my the rule of four for sizing the number of rollback segments
Thanx
Stefano Received on Fri Mar 02 2001 - 16:26:47 CST