On Fri, Aug 08, 2003 at 01:59:49AM +0100, Toad wrote:
> How will this limit output bandwidth, on even a minute by minute basis?

It is true that this approach has a slower feedback-loop than
reimplementing a Quality of Service TCP/IP stack within Freenet, as you
seem determined to do - but it is the natural way for Freenet to reduce 
its bandwidth usage (namely, not to take on more work when it is 
overloaded), rather than taking on more work - but just doing it more 
slowly.  Does it guarantee that if someone gets out Ethereal and starts 
to collect statistics that Freenet won't occasionally exceed the 
bandwidth allocation?  No, it doesn't.  Will it mean that Freenet tries 
to stick to the bandwidth allocation using reasonable means - yes it 
does - and I suspect that will be enough for 99.9% of Freenet's users.

> Users will regularly, and as part of the network's normal behaviour,
> have their link saturated by unluckily having a few downloads from
> people with fast links.

Only if the next slowest person in the reply-chain is faster than their
total upstream allocated bandwidth.  I am sure you can agree that this
will be rare unless their internet connection involves smoke signals.

> Since we want people to run freenet 24x7 in the
> background, in parallel to their regular browsing, this is a bad thing.

For those subscribed to SmokeSignalLink - yes, you may be right.  My
apologies to Freenet's isolated 17th century Navajo users.

> How will you limit incoming transfers?

By rejecting the requests that lead to incoming transfers in the first 
place.

Ian.

-- 
Ian Clarke                                                  [EMAIL PROTECTED]
Coordinator, The Freenet Project              http://freenetproject.org/
Weblog                               http://slashdot.org/~sanity/journal

Attachment: pgp00000.pgp
Description: PGP signature

Reply via email to