Migration to new SourceForge URL scheme part 2, SFE and some statistics

[Wesley Shields]

(Trimming CC line...)

On Wed, Sep 02, 2009 at 08:26:33PM +0400, Dmitry Marakasov wrote:
> * Paul Schmehl (pschmehl_lists at tx.rr.com) wrote:
> 
> > > In my case, over a few retries, switch is the fastest, followed by
> > > surfnet. heatnet is only somewhere between 500-700K.
> > > dfn, garr: and ovh fail.
> > 
> > It might be way too much work for very little benefit, but network latencies 
> > being what they are, perhaps there should be a routine that runs periodically 
> > and adjusts the list according to some connectivity parameters?  (Yeah, I know, 
> > easy for me to say.  I don't have to write the code.)
> 
> That's really too much work for a little benefit. The only thing
> we really want from mirrors is fetchability, and as soon as we have
> multiple mirrors that's achieved. Speed is a different issue and,
> as my survey shows, mirrors can't be sorted once and forever to
> satisfy the whole world, so if you feel like downloads are slow,
> just add your favorte mirror into make.conf, like I did a long time
> ago. The utility to do it automatically would be useful though, and
> actually you can write one. Actually, something even more clever
> could be written, similar to RANDOMIZE_MASTER_SITES, but based on
> fetch speed feedback, but that'd be an overcomplication if you ask me.

There is ports-mgmt/fastest_sites which sorts based upon round-trip time
for the TCP handshake to complete. It's not accurate for download speeds
but it provides a rough approximation for minimal effort.

-- WXS