INN-base feeder

Kenichi Okada okada at opaopa.org
Sat Jan 25 09:32:46 UTC 2003


Hi,

In the message "<20030125085115.GX2617 at virgin.gazeta.pl>"
Przemys=B3aw Maciuszko <sal at agora.pl> wrote:

> > I think your disk I/O is a bottleneck. If your CNFS is the half of your
> > main memory and your history is very small, then your can access them
> > very fast. It is an effect of kernel cache.

> Oh. That's the point....

> > In our server, the sum of CNFS and history is 1,1GB.
> > The main memory is 2GB.

> I have only 1GB memory... And I don't think 512MB will last for longer
> than a moment ;) My measurments show it will last for a 30sec :)
> Will it be enough?

If you do not mind that a lifetime of binary articles is short,
you can put the arcitles on a different buffer.
So, you should have asymmetric CNFS buffers for each type of articles.

Here's storage.conf.

# regional newsgroups
method cnfs {
        newsgroups: fj.*,jp.*,tnn.*,iij.*,pin.*,japan.*,okinawa.*,etl.*,kan=
sai.*,nara.*,ikedaken.*,yajima.*,lab3.*,tsuda.*,doshita.*,lab6.*,ishida.*,k=
ambayashi.*,satoken.*,yuasa.*,minoh.*,is.*,isle.*,imel.*,kuee.*,kuec.*,kuam=
p.*,kudpc.*,jccs.*,ku.*,nihon.*,feedmania.*,asatau.*,jape-n.*,jlug.*,kgk.*,=
comp.*,gnu.*,news.*,linux.*
        class: 1
        options: JAPAN
}=20

# binaries
method cnfs {
        newsgroups: *
        size: 30000
        class: 3
        options: BIN
}

# other
method cnfs {
        newsgroups: *
        class: 2
        options: OTHER
}



> So maybe put this CNFS buffer and history into ramdisk?

I use ramfs of Linux, and our CNFS buffers and our history
(only history.dir, history.hash and history.index) are on the ramfs.

Regards,

--=20
Kenichi Okada
mailto:okada at opaopa.org



More information about the inn-workers mailing list