| 1 |
On Friday 11 April 2003 02:01, Matt Thrailkill wrote: |
| 2 |
> > Besides building against GRP would mean a) maintaining a grp chroot |
| 3 |
> > (reasonable) and b) building every package twice - once for myself, once |
| 4 |
for |
| 5 |
> > grp. (No ccache because my cflags are different from grp's). So twice the |
| 6 |
> > build time. Ugh. |
| 7 |
> |
| 8 |
> Curious though, how much time is spent trying to build such and such an |
| 9 |
ebuild until you actually get it right and then finally compile something |
| 10 |
that works and installs fine on your box? |
| 11 |
|
| 12 |
Well ok, 'twice' isn't accurate. But it still does mean another full build of |
| 13 |
everything I commit, which is a lot by the same calculation that says it |
| 14 |
would take dedicated server(s) to do this. And we don't all have server-class |
| 15 |
development machines at home. The remote developers'-build-server is much |
| 16 |
better. |
| 17 |
|
| 18 |
> Do you think the onus is more on finding machines to compile it all with or |
| 19 |
machines to host it all with? I would think that some hotrodder overclocker |
| 20 |
types, the kind that run Gentoo cause they like watching gcc scrolling in an |
| 21 |
xterm, would be able to get an impressive distcc farm built out of the boxes |
| 22 |
they personally own. Ibiblio probably wouldn't be happy hosting binaries |
| 23 |
though. |
| 24 |
|
| 25 |
The hotrodders are ruled out for the same reason a p2p network is - we |
| 26 |
wouldn't be able to trust them (maybe a few we know, but not just anyone who |
| 27 |
has gentoo and a fast box). We need a server we can control and make sure is |
| 28 |
secure. |
| 29 |
|
| 30 |
If we have a lot of uptodate GRP packages, usage will spread very very rapidly |
| 31 |
and downloads will rival or perhaps pass distfile downloads (remember that |
| 32 |
only a part of the latter goes through our mirrors, some still download from |
| 33 |
packages' homesites). So even if all our existing mirrors ageed to host this |
| 34 |
too, it would more than double the bandwidth we'd use up. It would also |
| 35 |
multiply many times the amount of spaec we take on a mirror, because every |
| 36 |
package would exist for a geat many archs. |
| 37 |
|
| 38 |
|
| 39 |
-- |
| 40 |
Dan Armak |
| 41 |
Gentoo Linux developer (KDE) |
| 42 |
Matan, Israel |
| 43 |
Public GPG key: http://cvs.gentoo.org/~danarmak/danarmak-gpg-public.key |
Archives