On Sun, Feb 22, 2015 at 05:18:24PM GMT, Stuart Henderson wrote:
> On 2015-02-22, Raf Czlonka <rczlo...@gmail.com> wrote:
> > On Sun, Feb 22, 2015 at 12:43:57PM GMT, Stuart Henderson wrote:
> >> On 2015-02-21, Raf Czlonka <rczlo...@gmail.com> wrote:
> >> > Alternatively, you can use CVSync[0] to get the code.
> >>
> >> That is likely going to be worse on a slow/unstable link as you
> >> are fetching the entire repository history (~2GB rather than ~750MB),
> >> not just updating the checked-out tree.
> >
> > One would think that. However, CVSync + local 'cvs up' is much faster
> > and more reliable than "remote" 'cvs up'.
> >
> > I'm using it daily so this is not speculative - I'm writing from my own
> > experience. Also, I had very similar issues as the OP.
> >
> > Cheers,
> >
> > Raf
> >
> > P.S. I'm use CVSync with your mirror :^)
> >
> >
>
> Interesting.. I stopped running local cvsync to my server at home
> a while ago, as 'cvs up' from my mirror (over adsl) was faster, not
> even taking the extra cvsync time into account ;)
>
I double-checked and 'src' and 'xenocara' on their own are, indeed,
faster with 'cvs up' from a remote server so I stand (partially)
corrected - admittedly, previously I had used another, and as it appears
slower, mirror before switching to 'cvsync' and more "local" mirror :^)
Results are as follows with 'cvs up' from a local disk:
current src 8.83s user 21.58s system 8% cpu 6:03.74 total
current xenocara 2.94s user 6.81s system 4% cpu 3:25.79 total
vs. 'cvs up' directly from your mirror:
current src 5.12s user 12.46s system 11% cpu 2:34.07 total
current xenocara 1.54s user 3.82s system 7% cpu 1:12.62 total
'ports', however, is a totally different matter and it varies from:
current ports 8.63s user 40.46s system 3% cpu 21:49.16 total
to:
cvs [update aborted]: end of file from server (consult above messages if any)
with 'cvs up' from a remote mirror and:
current ports 11.28s user 43.68s system 7% cpu 12:18.71 total
with 'cvs up' from a local filesystem.
All in all, 'cvsync' (mirroring the whole repo takes ca. 4 min 30 sec) +
'cvs up' for all three ('src', 'xenocara' and 'ports') is marginally
(negligible) slower but, from my own experience, much more reliable.
I am sure 'cvs up' from a local filesystem could be sped up a bit if the
local mirror and the source tree were not on the same disk - these
results are from my single-drive laptop.
Regards,
Raf
