NNSquad - Network Neutrality Squad

NNSquad Home Page

NNSquad Mailing List Information

 


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[ NNSquad ] Re: Comcast announces tests of "protocol agnostic network management"


One of the problems, and why you would want to apply QOS rules to a
user when you are throttling them (and even when you are not!).

After all, although I'd love the utopia where the user's apps mark the
QoS packets right, anything involving changing end-hosts is pretty
much a non-starter (eg, where is ECN's universal deployment?)

So if you are throttling a user for being disruptive to other users,
you may still want to do DPI, or at least some heuristics to
prioritize how you handle the throttled user's traffic.    Why should
Mom's VoIP have to suffer because Junior is trying to upload "Indiana
Jones and the Latest Shiny MacGuffin"?!?


But the problem is even worse than that.  The EXISTING access devices
(cable modems, DSL routers) are often criminally poorly designed: They
don't support any form of fair queuing or active queue management, and
worse, are often horribly overbuffered (1-4+ seconds of buffer).

As a result, even if the network is UNCONGESTED for everyone else,
within a subscriber's houshold, Billy Junior's BitTorrent can disrupt
Mom's VoIP call, and just fielding the call and trying to explain to
Mom why her telephone service isn't working, and it isn't the ISP's
fault, costs $10.

Yet replacing the end-point boxes is probably unfeasible:  You're
talking $50/subscriber and even then, you're assuming that

a:  The replacement boxes won't still be criminally badly designed

b:  The users will be able to replace the boxes without a service
call.  If just 10% require a service callout, thats probably another
$20+/user.



Thus there is a strong argument that ISPs should do traffic managment
to keep a user's heavy flows from saturating their own access link.
Just reducing them to 95% of their theoretical peak by dropping
packets from heavy flows ("Remote Active Queue Management" sounds like
a good buzzword to me) would have an almost trivial effect on bulk
file transfer performance, but a huge benefit for users as now
BitTorrent or any other bulk flow would not affect the rest of their
traffic by filling up the send-queue.