Michael Widenius wrote:
>
>
> To hold 600 tables, you will only need about 600*x+5 file descriptors.
>
> Here x is between 1 and 2, depending on how many 'unique' tables you
> are using; If you open one table 600 times, x is 1; If you open 600
> different tables, x is 2.
>
> Regards,
> Monty
>
>
My application is a logging process that at regular time intervals sequentially
reads
acquisition values from 600 hardware devices (ADC etc...) and inserts the
values into 600 different tables, one for each device. I assume that in this
situation
the table cache will not be of any use, since the requested table will never be in
the
cache if the cache size is < 600. One could of course organise the data
differently,
but the person who originally wrote the application is no longer working for us so
I thought that the simplest way to speed up the program would be to increase the
table cache to hold all the tables.
Since each table is represented by three files, I (incorrectly) assumed that I
would
need 3x600 filedescriptors. From the above I make the new estimate 600x2+5.
From the manual I also note that locking the tables would increase insert speed,
but I also note that locking isn't working on the Linux platform. Has this changed
with the 2.2 kernel ?
--leif
---------------------------------------------------------
Leif Thuresson Tel: +46-18-4713860
Uppsala University Fax: +46-18-4713833
The Svedberg Laboratory E-mail: leif.thuresson@stripped
P.O.Box 533
S-751 21 Uppsala
Sweden
