Re: HELP: RAID-5 Killing Performance

There are supposed to be some operating systems where the call to time() has a significant cost, but on Unix boxes it seems to be very cheap.

The worst overhead I ever saw was about 10%, but this was on a totally abysmal application that did everything with single row fetches - and then did literally millions of fetches.

On most systems that I have seen the overhead appeared to be less than 1%.

My theory is this:

    If the overhead is enough to make you worried,     you need to tune the database. If you need to     tune the database you need to have timed_stats     on. If you don't need to tune the database you     can afford to have it timed_stats on anyway.

There's a famous law of human behaviour that sounds similar, but I can't remember which it is.

--

Jonathan Lewis
Yet another Oracle-related web site: http://www.jlcomp.demon.co.uk

Connor McDonald wrote in message <37DF762C.255_at_yahoo.com>...
>
>On the topic of timed_statistics, I have this:
>
>(From the Oracle doco)
>
>"If TIMED_STATISTICS is FALSE, the statistics related to time are always
>zero and the server can avoid the overhead of requesting the time from
>the operating system. To turn on statistics, set the value to TRUE.
>Normally, TIMED_STATISTICS should be FALSE. "
>
>(From the Metalink site, in the Oracle document on ultbstat/utlestat
>
>"Setting TIMED_STATISTICS to TRUE adds only a very small overhead to
>database operations - it is advisable to keep this set at all times"
>
Received on Wed Sep 15 1999 - 13:49:27 CDT