Re:The lost Van Jacobson paper that could save the Internet
by xtacy on Friday December 30, 2011 @12:47PM
(#3408360)
Interesting. I've been looking into the Stephen Savage paper faking-acks part as well for a while now, especially in the context of hosted services in a shared infrastructure (cloud!). My hunch is that it can cause a lot of problems, but I am yet to verify. :)
Re:The lost Van Jacobson paper that could save the Internet
by eternalban on Friday December 30, 2011 @11:47AM
(#3408163)
Am interested in your opinion on using Bayesian or Fuzzy approach to this system control problem.
Re:The lost Van Jacobson paper that could save the Internet
by eternalban on Friday December 30, 2011 @03:29PM
(#3408971)
http://privatewww.essex.ac.uk/~fleum/csvt.pdf for example.
The Stefan Savage paper you refer to wasn't so much about ignoring congestion signals as deliberately faking reception of data that hadn't actually been received. With things like duplicate acks or optimistic acking, you could cause the sender to open their window size to whatever you want.
http://cseweb.ucsd.edu/~savage/papers/CCR99.pdf
Back in 1999, I was working on integrating these ideas into a router product. It would be software in your home gateway that would keep a cache of the last observed window size for each pair of hosts for a few seconds.
If you started another TCP connection to the same host within that short interval, the router would spoof both sides to increase the window size to the previous value / 2 and then let additive increase take over from there. This bypasses slow start but the current available bandwidth is likely to be very close to the estimate within that short range and hosts didn't cache window size, even when making multiple connections in a short period (e.g., HTTP 1.0)
I archived this paper and a bunch more here. Prepare to lose an afternoon if the history of TCP/IP interests you.
http://root.org/ip-development/