>> Clearly it _does_ improve his network performance, and there is
>> something very wrong with exponential backoff on the Internet.
>
> I am not following. Are you suggesting that because it improves
> his performance he should misbehave at the expense of other users?
Yes... and they can do the same. There are too many holes in the
rules already: UDP, short-lived connections, user behavior...
>> Congestion is not a temporary problem that can be fixed in such
>> a simple manner.
>
> People have studied this algorithms for years; written a lot
> of papers. You are making a blanket statement but i will give
> you the benefit of doubt. I would like to hear your solution.
I don't believe there is a solution. The current rules are obviously
not working well, and can sometimes make the problem worse for everybody.
Considering such failure, it would be nice to reduce the annoyance.
>> Even if every client does the backoff (hah!)
>> there will be plenty of packets to which it does not apply.
>
> Most of the internet traffic is TCP which conforms.
Does it? You assume that TCP is used for bulk data transfers
that won't be restarted by a human. Add the human to your math.
> UDP packets mostly dont(i guess thats why realvideo runs so well).
Gee.
> This is not about the user perceiving delays; it is about avoiding
> congestion collapse.
Fine, how do you stop the human? How do you keep people from sending
lots of tiny images in parallel connections?
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/faq.html