lkml.org 
[lkml]   [1999]   [May]   [28]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [patch] TCP/IP delacks disabled with MPI
On Mon, May 24, 1999 at 06:06:00PM +0200, Andrea Arcangeli wrote:
> >Quick ack mode is a existing mechanism that could be extended to fix it
> >better.
>
> Could you explain better your way?
>
> quickack is just a bit that when is set then delacks are forbidden. When
> will you set the quickack bit?
>
> Right now we use it to force the first ack after the three way handshake
> because at that time the sender has cwnd = 1. So with quickack we avoid
> having to wait ato to expire before allowing the sender to give us more
> data.
>
> Right now we use it also when we receive retransmitted or out of order
> packets, in order to advertise the sender aboute our current rcv_nxt.

In short it is always set when we have a strong cue that the sender is in
slow start. The problem is that Linux doesn't send acks quickly enough to
get the sender quickly out of slow start when it sends lots of small packets.
Quickack does in principle the right thing, but it works for too few acks (1).
My proposal is to extend the quickack state to more acks, so that it speeds
up the many small packets too (instead of only the single small packet HTTP
case). Only thing it has to make sure is that the estimate about the
sender's slow start makes sense - extended quickack shouldn't trigger for
a fast retransmit.


-andi




--
This is like TV. I don't like TV.

-
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/

\
 
 \ /
  Last update: 2005-03-22 13:51    [W:0.050 / U:0.488 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site