Being reaqsonable, my idea would never have worked...
But I have to say that load average is, actually, a _really_ bad way of
testing load on a box.
If you're running the webserver, then I'd say apache's resource usage is
probably the best...
> -----Original Message-----
> From: Danil Lavrentyuk [SMTP:[EMAIL PROTECTED]]
> Sent: Tuesday, May 29, 2001 3:57 PM
> To: '[EMAIL PROTECTED]'; Briggs, Gary
> Subject: RE: sleep when system is heavily loaded
>
> [ On Tue, 29 May 2001, Briggs, Gary wrote: ]
>
> BG> Have you tried writing an independant daemon to do this?
> BG>
> BG> Just send indexer SIGSTOPs whenever the load average goes above
> whatever,
> BG> then SIGCONTs whenever it drops again. This would have the added
> advantage
> BG> of being almost no resource usage, and the indexer would never need to
> BG> know...
>
> Think this way is not so good.
> I afraid the next sequence of events could happen:
>
> 1) An indexer just started to reseiving a page to index.
> 2) A daemon described above checks load average and thinks it is a time to
> send SIGSTOP to the indexer.
> 3) The indexer is sigstopped.
> 4) ...some time have passed...
> 5) The daemon sends SIGCONT to the indexer.
> 6) The awakened indexer checks timeout and thinks "Too long waiting for
> this
> page." And stops the receiving.
>
> Tell me wy this can't happen?
>
> It would be good if an indexer go sleep itself in *safe* moment of time.
>
> BG> But for reference, why use load average? It's usually a bad measure of
> BG> what's going on in a system, and unreliable as hell.
> BG>
> BG> If you're running the db on the same box as the indexer, you may
> BG> accidentally throw the whole system out of kinter because an i/o bound
> DB
> BG> can increase the load average more than is entirely natural, so you'd
> get an
> BG> ugly cycle going on...
> BG>
> BG> Depending on what the box is doing, if it's only a web server then why
> not
> BG> just look at Apache's resource usage?
> BG>
> BG> And so on and so forth.
>
> Thank you but all your ideas are no reliable no more then load average
> usage.
> All these criteria have their strongnesses and weaknesses.
>
> ----------------
> Danil Lavrentyuk
> Communiware.net
> Programmer
>
___________________________________________
If you want to unsubscribe send "unsubscribe general"
to [EMAIL PROTECTED]
