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

Re: [E-devel] Requirements for a mirror of anon CVS

On Wed, 29 Mar 2006 07:30:02 -0800 "Blake B." <shadoi@nanovoid.com> babbled:

> On Mar 28, 2006, at 7:03 PM, Carsten Haitzler (The Rasterman) wrote:
> > On Tue, 28 Mar 2006 18:55:09 -0800 Ben Ford <ben@kalifornia.com>  
> > babbled:
> >
> >> Hash: SHA1
> >>
> >> How much bandwidth would it take to run a mirror?  My university LUG
> >> has plenty of disk space, but I'd need to get permission from IT
> >> before we start pushing out gigs and gigs of traffic.
> >
> > bargain on many GB/day - thinktux was doing 25GB/day.
> >
> It was also the only AND preferred mirror.  I think we could  

it was - but there was still sf.net which imho was still taking the lions share of the anoncvs load - but because we never have been able to gather any anoncvs stats - it's all conjecture. you should not "expect" anything. be prepared to have a lot of bandwidth load if you run an anoncvs mirror and don't get upset if you have quotas and they get blown. :)

> distribute that load fairly well with 4 or 5 mirrors.  A good way to  
> load balance them would be awesome.  I don't think IPVS would work,  

no it won't. ipvs will still put all the bw through the LD.

> and DNS round-robin is too tricky with possible synchronization  
> issues.  Anyone have any ideas?

we can't do dns round-robin anyway because easydns doesn't handle that. also u need to be connecting to the same anoncvs mirror each time for consistency - the mirror(s) will rsync at different times ... in fact they SHOULD have staggered rsyncs - not all rsync "on the hour" or something. this is why we need to at least for now

1. set up an "anoncvs mirror package" which is all the configuration you need to set it up - and maybe software too.

2. i want to track statistics. if we can't track stats we keep saying "i have no idea" to people who ask how much it takes to run an anoncvs server - and we have no idea how heavily cvs is really used. is it just a very vocal small group of people or what? we need a way to gather stats (how many GB/day/hr etc. - how many connections, how many unique ip's - where do the connections come from (usa, europe, asia etc.). this will also let us know where the need for anoncvs servesr is greates - eg we may need 2 or 3 in usa, 1 or 2 in europe, only 1 in asia etc.

> -Blake

------------- Codito, ergo sum - "I code, therefore I am" --------------
The Rasterman (Carsten Haitzler)    raster@rasterman.com
Tokyo, Japan (東京 日本)