Re: [alto] alto Digest, Vol 24, Issue 5

[Enrico Marocco <[e-mail]>]

On 10/13/10 5:21 AM, wangaijun wrote:
> In this charter, it is said, "... In any case, this WG will not
> propose standards on how congestion is signaled, remediated, or
> avoided, and will not deal with information representing
> instantaneous network state.  Such issues belong to other IETF areas
> and will be treated accordingly by the specific area."

That part of the charter -- that was discussed, understood and agreed
before the creation of the working group -- means exactly what it says:
congestion is a transport issue that varies on a different time scale
than the network topology ALTO is about. As such, it needs to be handled
at a different layer.

The problem with congestion is that it changes very quickly and is
subject to feedback loops. That means, in the case of distributed
applications, that avoiding a congested path is likely to cause
congestion on some alternative paths. Of course it is not an unsolvable
problem -- the Internet has been dealing with congestion since day one,
one may say it is part of the very nature of packet networks -- but is
one that needs to be addressed instantaneusly. An application protocol
like ALTO just cannot provide such frequent updates: transport protocols
do, and are constantly being enhanced to do it better (see the LEDBAT
and CONEX working groups for example).

This of course does not preclude an ALTO server to assign higher
priorities to networks that are better connected, i.e. whose connecting
links have, on average, higher available bandwidth. As long as the
average is estimated on a timescale that is comparable to the TTL of the
information the server distributes (i.e. days, possibly weeks or months).

-- 
Ciao,
Enrico