Yes, that blog post by Jim is what led me to investigate and write about this. The Stefan Savage paper you refer to wasn't so much about ignoring congestion si

Yes, that blog post by Jim is what led me to investigate and write about this. 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 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/ http://root.org/ip-development/