On Wed, 2002-09-04 at 12:53, Florin Andrei wrote:
> 
> That's odd, i'm using it precisely because it has a good INSERT
> performance, which corresponds to my tests. 
> But indeed, i didn't tested with a large number of connections, so this
> might be an avenue... 

It's probably worth considering if performance becomes a serious issue.

> > The only thing that I wonder about is that your load is mostly in the
> > "system".  You might get better performance by simply adjusting your
> > read-ahead and write-cache settings.  Try to figure out how *much* disk
> > activity is being generated by the server using "vmstat".
> 
> [root@xxxx irq]# vmstat 1
>    procs                      memory    swap          io    
> system         cpu
>  r  b  w   swpd   free   buff  cache  si  so    bi    bo   in    cs  us 
> sy  id
>  1  0  0   1152  11676      4 811776   0   0     0     0  175  1260   9 
> 34  57
>  0  0  0   1152  10700      4 811796   0   0     0     0  163  1047   6 
> 32  63
...
> Seems like there's not much disk IO, but lots of context switches.
> Could it be that the system spends too much time just switching between
> threads?

Even at the heaviest use, it doesn't look like there's much of a problem
here.  The system is writing to disk very infrequently, and well over
half of the CPU time is spent idle.  Is the performance of the system
really a problem, or are you just curious about the CPU utilization?

Your context switches don't seem unusually frequent, nor does the system
CPU use seem unusually high.  Compared to a couple of busy servers here,
it looks fine.

If you're curious about context switches, you might be able to
streamline the process some.  I assume that you wrote the "custom
daemon" sending this data.  Maybe you could write a daemon that listened
on the machine running the SQL server for incoming requests, formatted
them in a SQL statement and fed them to the server.  It could handle
connections in an array of fd's, using select() to poll for incoming
data.  You'd have one SQL server, and one daemon;  the process should be
extremely efficient.  However, given the stats, I wouldn't guess you'd
get much ROI.




-- 
redhat-list mailing list
unsubscribe mailto:[EMAIL PROTECTED]?subject=unsubscribe
https://listman.redhat.com/mailman/listinfo/redhat-list

Reply via email to