Received: by oss.sgi.com id ; Fri, 5 Jan 2001 06:09:41 -0800 Received: from colin.muc.de ([193.149.48.1]:43536 "HELO colin.muc.de") by oss.sgi.com with SMTP id ; Fri, 5 Jan 2001 06:09:14 -0800 Received: by colin.muc.de id <140550-3>; Fri, 5 Jan 2001 15:08:53 +0100 Message-ID: <20010105150849.45985@colin.muc.de> From: Andi Kleen To: Karl Jonas Cc: netdev@oss.sgi.com Subject: Re: looking for info about struct tcp_opt References: <3A55CF54.8C890679@ccrle.nec.de> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Mailer: Mutt 0.88e In-Reply-To: <3A55CF54.8C890679@ccrle.nec.de>; from Karl Jonas on Fri, Jan 05, 2001 at 02:35:54PM +0100 Date: Fri, 5 Jan 2001 15:08:49 +0100 Sender: owner-netdev@oss.sgi.com Precedence: bulk Return-Path: X-Orcpt: rfc822;netdev-outgoing Content-Length: 1022 Lines: 36 On Fri, Jan 05, 2001 at 02:35:54PM +0100, Karl Jonas wrote: > WHAT IS THE UNIT OF SRTT? > According to its declaration in sock.h, it is the 'smoothed rtt << 3' . > Does this mean, that the value '8', which i observed, corresponds to a > round trip time of about 1 ms ? In the linux kernel time is usually counted in jiffies (=timer interrupts), on i386 that's 10ms, e.g. on alpha it is ~1ms. Unit is jiffies<<3. > > WHAT IS THE UNIT OF SND_CWND? > In my trace, snd_cwnd initialises with 3 and increases by 1 for each > packet > sent (acked?), and slower after, corresponding perfectly to slow start / acked > congestion avoidance. But what is the real size of the congestion window > (maybe snd_cwnd * mss ?) ? It's packets. 2.4 switched to counting it in bytes, because the packet counting has some problems. factor is *average of your packet size. > > WHAT IS THE RTO UNIT? > I see a rto of 20. This seems to be quite short if it was 20 ms. Does > linux > count in 10's of ms here? Yes, in jiffies. -Andi