[PATCH 08/29] Honour initial RTT estimate

[PATCH 08/29] Honour initial RTT estimate

[Arnaldo Carvalho de Melo]

This is a small optimisation which improves the accuracy of TX
RTT sampling when an initial RTT sample (e.g. from the intial
Request/Response exchange) is available.

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Acked-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
---
 net/dccp/ccids/ccid3.c |   13 ++++++-------
 1 files changed, 6 insertions(+), 7 deletions(-)

diff --git a/net/dccp/ccids/ccid3.c b/net/dccp/ccids/ccid3.c
index a20b636..9d661e4 100644
--- a/net/dccp/ccids/ccid3.c
+++ b/net/dccp/ccids/ccid3.c
@@ -469,28 +469,29 @@ static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
 		else				       /* can not exceed 100% */
 			hctx->ccid3hctx_p = 1000000 / pinv;
 
-		skb_get_timestamp(skb, &now);
-
 		/*
 		 * Calculate new round trip sample as per [RFC 3448, 4.3] by
 		 *	R_sample  =  (now - t_recvdata) - t_elapsed
 		 */
+		skb_get_timestamp(skb, &now);
 		r_sample = dccp_sample_rtt(sk, &now, &packet->dccphtx_tstamp);
 
 		/*
-		 * Update RTT estimate by
-		 * If (No feedback recv)
+		 * Update RTT estimate (honours RTT from SYN exchange):
+		 * If (No RTT sample so far)
 		 *    R = R_sample;
 		 * Else
 		 *    R = q * R + (1 - q) * R_sample;
 		 *
 		 * q is a constant, RFC 3448 recomments 0.9
 		 */
+		hctx->ccid3hctx_rtt = hctx->ccid3hctx_rtt = 0 ?  r_sample
+				    : (9 * hctx->ccid3hctx_rtt + r_sample) / 10;
+
 		if (hctx->ccid3hctx_state = TFRC_SSTATE_NO_FBACK) {
 			/*
 			 * Larger Initial Windows [RFC 4342, sec. 5]
 			 */
-			hctx->ccid3hctx_rtt  = r_sample;
 			hctx->ccid3hctx_x    = rfc3390_initial_rate(sk);
 			hctx->ccid3hctx_t_ld = now;
 
@@ -504,8 +505,6 @@ static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
 
 			ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
 		} else {
-			hctx->ccid3hctx_rtt = (9 * hctx->ccid3hctx_rtt +
-						   r_sample) / 10;
 
 			/* Update sending rate (step 4 of [RFC 3448, 4.3]) */
 			if (hctx->ccid3hctx_p > 0)
-- 
1.5.0.6

Re: [PATCH 08/29] Honour initial RTT estimate

[David Miller]

From: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Date: Thu, 12 Apr 2007 18:15:05 -0300

> This is a small optimisation which improves the accuracy of TX
> RTT sampling when an initial RTT sample (e.g. from the intial
> Request/Response exchange) is available.
> 
> Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
> Acked-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
> Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>

Ok, even more...

I'm not pulling this stuff in, you guys have to work out a way
to implement this without killing performance on the rest of
the system.

Thanks.

Re: [PATCH 08/29] Honour initial RTT estimate

[Arnaldo Carvalho de Melo]

On 4/12/07, David Miller <davem@davemloft.net> wrote:
> From: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
> Date: Thu, 12 Apr 2007 18:15:05 -0300
>
> > This is a small optimisation which improves the accuracy of TX
> > RTT sampling when an initial RTT sample (e.g. from the intial
> > Request/Response exchange) is available.
> >
> > Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
> > Acked-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
> > Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
>
> Ok, even more...
>
> I'm not pulling this stuff in, you guys have to work out a way
> to implement this without killing performance on the rest of
> the system.

I'm removing this one as well.

- Arnaldo

Re: [PATCH 08/29] Honour initial RTT estimate

[Gerrit Renker]

|  > This is a small optimisation which improves the accuracy of TX
|  > RTT sampling when an initial RTT sample (e.g. from the intial
|  > Request/Response exchange) is available.
<snip>
|  
|  Ok, even more...
|  
|  I'm not pulling this stuff in, you guys have to work out a way
|  to implement this without killing performance on the rest of
|  the system.
|  
I take it you are referring to the moved skb_get_timestamp in this patch?

As part of of reverting the RX/TX timestamp patches, this will go back
to either dccp_gettimestamp() or do_gettimeofday() (will discuss with Arnaldo)
- as will be all other patches that depend on the RX/TX timestamps.

When that is removed, I'd say that this patch is useful since it increases
the number of samples used to average the RTT.

I will post a summary of all changes to the list.