Re: access time performance problem

On Thursday 10 Oct 2002 9:51 am, Louis-Marie Croisez wrote:
> Sorry, bad url:
> use this instead:
> http://louis.croisez.free.fr/capture1.jpg
>
>
> > Thanx for your response Scott.
> > In order to illustrate my problem, here is a capture of an Excel graph
> > showing the access peaks I get:
> > http://louis.croisez.free.fr/download/capture1.jpg
> > The test program makes updates of a table (on 1000 records randomly)
> > every 50ms. Approximately, every about 3 seconds, the update process is
> > hanged for about 800ms (these data are measured on the Compaq. On the
> > IBM, the results are worse).
> > This is a great problem for us.
> > Could you give me some help to correct that ?

I'm no expert on fs issues, but I do remember reading about bursts of write
activity occuring with ext3 (which you mentioned using in a previous mail,
and Scott remarked upon). IIRC it was something to do with ext3 buffering for
a period and then performing a bunch of writes at the same time. This looks
like a plausible candidate for your problem. The bad news is you'll have to
do some googling, since I can't remember where I heard about it. The good
news is that I seem to remember tuning that would help.

You might find vmstat/iostat useful to identify where the slowdown is
occuring.

- Richard Huxton