Re: TCP ex Machina

I think that the conversation is getting ramped up now ;^)
|
My thoughts on the paper are mixed.  I believe that centralized server calculated CC is the wrong direction for the future.  IMHO there are simply to many clients in network conditions that are too diverse, and too rapidly changing.  One size doesn't fit all.  I think that the client should provide the congestion control algorithm.

So the paper takes a nice step in this direction.  The client provides the network conditions to the server, then the server sends the CC algorithm to the client.  This still seems like an overload of work for the server, but, it is necessary because of the ex Machina part of the paper.

I think that the client should control the CC algorithm.  This takes care of a host of problems.  The client is in a better position to understand it's ever changing network conditions than the server.

Also I think that relying on things like packet inter-arrival time for CC is a foolish mistake.  There was a masters project that did this at my University.  TCP Santa Barbara.  It performed well in simulation with other TCP Santa Barbaras.  However, it could not play nice with VJCCs, nor, could it survive in the wild.  One has to be careful with parameters like inter-arrival time.  They are interesting, but, can only be relied upon in specific network conditions.

My final thought is that this new Yahoo mail interface should burn in hell.

...Daniel



________________________________
 From: Detlef Bosau <[email protected]>
To: [email protected] 
Sent: Thursday, September 19, 2013 1:04 PM
Subject: Re: [e2e] TCP ex Machina
 

Unfortunately, the discussion I wanted to initialte did not really start
perhaps due to the difficulties with some posts?

Nevertheless, I would appreciate any comment on my remarks.


Thanks,

Detlef