[Top] [All Lists]

Re: Karn's rule in Linux TCP

To: ak@xxxxxxx (Andi Kleen)
Subject: Re: Karn's rule in Linux TCP
From: kuznet@xxxxxxxxxxxxx
Date: Wed, 8 Nov 2000 20:19:08 +0300 (MSK)
Cc: davem@xxxxxxxxxx, netdev@xxxxxxxxxxx
In-reply-to: <20001106220115.A26879@xxxxxxxxxxxxxxxxxxx> from "Andi Kleen" at Nov 6, 0 10:01:15 pm
Sender: owner-netdev@xxxxxxxxxxx

> When the initial RTT estimate is too low then there will be no segments going 
> through
> without any retransmits. On a connection with no timestamps all RTT estimates 
> for
> retransmitted packets are ignored due to Karn's rule. Unfortunately this 
> means that
> when the initial RTT is too low it'll never get a new estimate, because all
> packets that arrive at the other end were already retransmitted. The 
> connection
> gets lots of faulty retransmits forever. Oops.

No oops. This RFC statement simply means that connections with rtt>3sec
are not going to work with TCP.

Partially, it was solved old days in Linux (specially for ultra-slow AX.25
links) with static rtt option on routes.

> Any suggestions on that?  Any other ideas? 

Right idea, in theory. But Andi, where did you find connections with rtt>3sec?
The worst link, which I ever saw was faulty 2400baud line and it had
rtt a bit lower. 8)


<Prev in Thread] Current Thread [Next in Thread>