[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Some real documentation on tuning

On 25/11/2002, Darren Reed <avalon_(_at_)_coombs_(_dot_)_anu_(_dot_)_edu_(_dot_)_au> wrote Cc misc_(_at_)_openbsd_(_dot_)_org:
> > sometimes it's required to turn on the brain and to read 'between the lines'.
> So why did "netstat -m" receive such poor treatment where your paper
> talks about NMBCLUSTERS ?  Non-zero values in Recv/Send-Q will impact
> the output of that.

>From the paper:
Let's have a look at the memory usage with {\tt netstat -m}. The stack has
to keep 85 clusters in KVM, somehow the application is processing data
either too fast (Send-Q) or too slow (Recv-Q). 
85/1048 mapped pages in use

So? Read it again. Combine stuff. Like with netstat -nfinet.
Anyway it's a better idea to use vmstat -m.

> > THINK about it, it doesnt help. No matter if arbitrary or "real measured"
> > values.
> But at least it adds some realism to what's reported.


> If developers don't want to support custom compiled kernels then it
> seems to me of little benefit extolling the virtues of doing this.

You can change e.g. nmbclust w/o recompile.

> > > Well if that's what you're arguing for, then I don't know if there's
> > > any good resource for that available for any BSD, any more.
> > 
> > Start in /usr/src/sys/kern
> That's not a very good reponse.

I could say that more often.

Philipp Buehler, aka fips | sysfive.com GmbH | BOfH | NUCH | <double-p> 

#1: Break the clue barrier!
#2: Already had buzzword confuseritis ? 

Visit your host, monkey.org