Re: timed statistics

In article <singerap.36.00179CEE_at_powergrid.electriciti.com>, singerap_at_powergrid.electriciti.com (Paul Singer) wrote:

> In article <D0G7FD.7An_at_ottnews.shl.com> mbelle_at_ottnews.shl.com (Marc
Bellefeuille) writes:  

> >I am wondering if when setting the init.ora parameter "timed_statistics"
> >to "TRUE" has a major impact on performance.
>
> I cannot tell you exacty how much slow it made thing. However, I can tell
> you it was quit noticable. I could not recomed this for production
 database.
> When I tund timed statistics on the users said every thing ran four times
> slower. This was on a Sequent system with 6 processors.
>

The method of getting the time differs for each O/S. It just depends on the machine architecture and how Oracle gets the time for that machine.

For us Digital fans, the time required to get the time on VAXs was on the dreadful side. However, it is a blindingly fast and simple LD instruction on the Alpha!

The Oracle Performance Tuning book (published by O'Reilly, not Oracle) recommends to leave timed_statistics on. I tend not to agree. It really is a platform-specific issue.

-- 
Jef
  "...I wouldn't normally do this kind of thing." -- Pet Shop Boys