Comments on the fastestmirror plugin

[Patrick O'Callaghan]

On Sat, 2010-03-13 at 23:16 -0500, Mail Lists wrote:
> On 03/13/2010 11:06 PM, Patrick O'Callaghan wrote:
> 
> > Sadly, downloading a "small" RPM is unlikely to give very reliable
> > results either. Due to TCP slow-start, a stable effective b/w may only
> > be reached after some 10's of kb have been downloaded.
> > 
> > This is not an easy problem to solve.
> > 
> > poc
> > 
> 
>  Perhaps instead of thinking about deterministic solution ..
> 
>  How about monte carlo samping the mirrors - some small number of draws
> (n)- choose best of those - then next update - do another MC sample -
> every update take n more draws -  after a while it should converge to a
> number of decent mirrors. In fact one could even break the update up
> into sub components - and run in parallel.

I like the idea of running (say) three dl's in parallel. If one server
is markedly slower than the others, discard it for the next time and
pick another one. The current ping-based test would be used only to get
the starting set of servers to try, so its deficiencies would tend to
self-correct in time.

poc