Re: [?] no logging mode

"gp" <gieppetto_at_tiscali.it> wrote in message news:bvb2vl$ri1$1_at_carabinieri.cs.interbusiness.it...
> do you think that i must increase log_buffer parameter ?
> (now it is 16 MB)
> thanks.
> Ciao.

Lord! The log_buffer should never, ever, ever need to be bigger than 1MB. Ever.

Well, except on those extremely rare occasions when maybe 3 or 4MB might be in order. But those are extremely rare, so I'd prefer to make your life easier by sticking with 'never, ever'.

Really. The log_buffer is (or ought to be) tiny, because redo is tiny. You're talking about a few hundred bytes at most being generated per row modification, and with in-built cleansing mechanisms such as the flush at 1/3rd full, every commit and before DBWR to keep the persistence space demands on the log buffer down to negligible levels. It's not like the buffer cache, that deals with blocks of data in the multi-thousands of bytes sizes, where blocks hang around for quite long times, where DBWR flushes comparatively rarely and so on. Buffer Cache big, log buffer tiny.

I suppose the real question is: what makes you think you might need a log buffer of 16MB, and what makes you do daft things like (attempt to) turn all tablespaces into nologging mode? It wouldn't be because you are suffering from log buffer space wait events or something, would it? Because those are not the cures for that particular problem.

Tell you what: tell us why you thought you needed to do these things first, and then we can have a go at explaining what the real cures are. But 16MB for a log buffer is way, way too much.

Regards
HJR

-- 
------------------------------------
Oracle insights at www.dizwell.com
------------------------------------