Re: package splitting and kernel-image
On Tue, May 15, 2001 at 05:36:06AM -0700, David Schleef wrote:
> On Tue, May 15, 2001 at 09:23:17PM +1000, Daniel Stone wrote:
> >
> > The point I'm making is that no mirrors here, nor anywhere, need the extra
> > stress of duplicated packages like this, and we really shouldn't be
> > stressing them.
>
> Where is this written in policy?
It's just commonsense. IMHO.
> While were at it, could you quantify the extra resources needed
> to have a duplicate copy of a packge on a mirror? Given that
> almost everyone downloading packages from a mirror will only
> download one of the duplicated binary packages, it seems to me
> that the added resources necessary are:
>
> - network bandwidth to download 1 binary image from master.d.o
> - disk space for 1 binary image
> - network bandwidth for downstream mirrors, if any
Yes, but when this is kernel-{image,headers} we're talking about, that's
still quite some load.
> Multiplied by N_ARCH, of course. But then, not every mirror
> mirrors every architecture or every package. If you really
> care about mirror stress, perhaps you could start publishing
> a list of daniel-stone-approved package and a script to allow
> mirrors to only mirror those packages. Make sure that my
> packages are not on it. (My packages, by the way, are probably
> less useful than an added kernel-image or ash and take up more
> resources.)
I don't have the time to build it on this crappy machine (mercifully, that
will change on Sunday), and somehow I don't think my ISDN would stand up to
the load of mirroring.
--
Daniel Stone
daniel@kabuki.openfridge.net
Reply to: