802.11 is a bad example.  It needs reliability because its losses
aren't well-correlated with congestion.

I'm not terribly concerned about TCP having reliability when used in
this context, although the stream-oriented nature of TCP is
unfortunate.  I don't want to go to a lot of effort adding reliability
to a UDP-based protocol, because I don't think it will add much (and
could hurt if done wrong).

Even if the overlay link protocol does have reliability, you still
need to shed load with finite-length queues (or push back) to avoid
congestion collapse/infinite queues.  I'm curious what literature
you're thinking off that argues otherwise.

Bruce


On Wed, Apr 8, 2009 at 8:27 AM, Henning Schulzrinne <[email protected]> 
wrote:
>
> On Apr 8, 2009, at 6:21 AM, Bruce Lowekamp wrote:
>
>> One of the (if not THE) fundamental decisions that makes the Internet
>> work is that reliability is end-to-end, not hop-by-hop.  Hops are
>> allowed to drop traffic due to congestion, and that is taken as
>> implicit feedback that there is congestion in the network.
>
> You may want to re-read the end-to-end argument paper. You do need
> end-to-end reliability to ensure reliability (since nodes in the middle can
> do bad things to packets), but performance often dictates "hop-by-hop",
> particularly if the "hop" is essentially the whole Internet. Since you like
> to use the notion of a link layer in P2P: this is the reason why 802.11 has
> link-layer reliability. By your argument, 802.11 should dispense with that.
>
>
>
>>
>>
>> The proposal I sent out drops traffic in two places.  First, each peer
>> maintains a limited queue for fragments and drops excess fragments
>> (the literal analog of drops due to congestion in routers).  Second,
>> the overlay link protocol is only semi-reliable (which I debated
>> having at all), with the assumption that on the Internet, loss is due
>> to congestion.  So we have two congestion signals to the peer, queue
>> length and link protocol drops.
>
> Unfortunately, that assumption is only somewhat true, particularly with
> wireless links.
>
>
>>
>>
>> Congestion needs to cause load shedding in a network (overlay or not)
>> or else it will collapse.  Arguing that we should do extra work for
>> hop-by-hop reliability makes no sense to me.
>
> Please consult the literature; this is a topic that has received more than
> its share of real measurement work, albeit ten and twenty years ago, when
> this topic was of greater practical interest.
>
>>
>>
>> Bruce
>>
>
>
_______________________________________________
P2PSIP mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/p2psip

Reply via email to