[bittorrent] Choking based on network closeness?
justin at street-vision.com
Tue Jul 12 12:52:03 EDT 2005
On 12 Jul 2005, at 17:34, Bill McGonigle wrote:
> On Jul 11, 2005, at 14:46, Justin Cormack wrote:
>> It came up on the list before, and there were some uses for it.
>> The main problem is that no one seems to have enough information
>> to measure closeness, unless you actually work for a large ISP or
>> run BGP.
> I've looked at a few different techniques that might apply -
> traceroute, pathchar, packet-tailgating. Assuming 'closeness' is
> only one decision criterion I'm not seeing a reason why traceroute
> wouldn't be sufficient (but tell me otherwise). You'd have to make
> some assumptions about how many hops are behind those * * * routers
> but the core internet routers seem to always respond properly, and
> that gets you most of the useful big-picture topology.
> There are certainly broken routers, some edge routers that won't
> honor the ICMP packets, and probably an OS that won't allow the
> user to send ICMP packets. But there is probably enough
> information most of the time to make a good guess. And, you'd
> always fall back to the 'fastest peer' algorithm in the absence of
> other data.
Even Linux requires root to send ICMP which is a bit of a pain
(though you can open a raw socket and drop privs). Would latency give
a good enough answer? It would penalize slow hosts too I suppose, but
it would be easier to measure...
More information about the BitTorrent