[Mep-dev] TCP Performance Issues

[Dave hartzell]

On Tue, Nov 25, 2008 at 1:57 AM, Phil Karn <karn@philkarn.net> wrote:

>> Dave Hartzell Wrote:
>>    TCP can perform well on wireless links, but the rest of the time it
>>    can drive you crazy!
>
> If it's not performing well on wireless links, then fix the link --
> don't tamper with TCP except to ensure that the window is big enough.

Sometimes, especially on shared RF bands, you cannot fix the link...

Most TCP stacks assume that loss = congestion, when it might just be a
short burst of interference.  The link may be uncongested, but yet TCP
backs off...

> I disagree. *Bad* TCPs perform poorly over long fat pipes. But unless
> the transfer is very short, there's no reason why a modern TCP can't
> fill the entire pipe.

The problem with "modern" TCP is all of the vendor implementations.
Sure, the code is the same within most of the Linux kernels, but the
"out of box" parameters are really messed up.  Time will be spent
tweaking the stack, with possibly minimal performance increases.

Supporting different hosts and different OSes for this project will be
interesting.  And there are TCP stacks suited towards wireless
links...but I'm not sure if those links are similar to ours, or
something like an 802.11 environment.