At 08:14 PM 9/18/2001 +0200, Leos Urban wrote:

>Values before parenthesis are for 8000 buffers, in parentheses for 100
>buffers.
>
>
>First connection after reboot (with empty filesystem cache)
>
>select sum(price) from inv_rows, inv_heads, companies
>where inv_rows.invcode=inv_heads.invcode and
>inv_heads.compid=companies.compid and
>companies.compname like 'A%'

First time Second time

>Elapsed time= 33.18 (33.82) sec Elapsed time= 16.68 (18.11) sec
>Cpu = 0.00 sec Cpu = 0.00 sec
>Buffers = 8000 (100) Buffers = 8000 (100)
>Reads = 7053 (20998) Reads = 0 (20916)
>Fetches = 1100224 Fetches = 1099269
>
>As I wrote, I need to know if difference times are normal, because for
>example, IB uses something by other way than Informix.

The cache is working, apparently. You notice that the number of
reads when the query is run first with the large cache is about
1/3 the number as with the small cache (7053 vs 20998). On the
second run, the larger cache did no reads, but still took 16.68
seconds.

If I could impose on you one more time, would you try this with
a cache of 4000? InterBase has known problems with larger cache
sizes and this is a very interesting test case.



Regards,

Ann
www.ibphoenix.com
We have answers.