[PATCH 5/5]: Implement rfc3448bis changes to feedback reception

[PATCH 5/5]: Implement rfc3448bis changes to feedback reception

[Gerrit Renker]

[CCID 3]: Implement rfc3448bis changes to feedback reception

This implements the algorithm to update the allowed sending rate X upon 
receiving feedback packets, as described in draft rfc3448bis, 4.2/4.3.

Some changes (use of Larger Initial Windows) were already present, this patch
adds the remaining ones. In particular:

 * use t_RTO to distinguish the case "initial feedback packet":
   as per earlier patch, it uses "state = NO_FBACK" to imply that 
     (a) either nofeedback timer expired (when t_RTO != 0) or
     (b) initial feedback packet         (when t_RTO = 0);
 * no reduction of sending rate afer nofeedback timer expiry when p = 0;
 * implements the clause ``if [...] not the first packet after a nofeedback 
   timer'' of draft rfc3448bis, 4.3.

Note: At some point of time this probably needs update wrt rfc3448bis, but see
      http://www.mail-archive.com/dccp%40ietf.org/msg00426.html

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
---
 net/dccp/ccids/ccid3.c |   64 ++++++++++++++++++++++++++-----------------------
 1 file changed, 34 insertions(+), 30 deletions(-)

--- a/net/dccp/ccids/ccid3.c
+++ b/net/dccp/ccids/ccid3.c
@@ -443,41 +443,45 @@ static void ccid3_hc_tx_packet_recv(stru
 	r_sample = dccp_sample_rtt(sk, ktime_delta(now, t_send));
 	hctx->ccid3hctx_rtt = tfrc_ewma(hctx->ccid3hctx_rtt, r_sample, 9);
 
+	/*
+	 *	Update allowed sending rate X as per draft rfc3448bis, 4.2/4.3
+	 */
 	if (hctx->ccid3hctx_state = TFRC_SSTATE_NO_FBACK) {
-		/*
-		 * Larger Initial Windows [RFC 4342, sec. 5]
-		 */
-		hctx->ccid3hctx_x    = rfc3390_initial_rate(sk);
-		hctx->ccid3hctx_t_ld = now;
-
-		ccid3_update_send_interval(hctx);
-
-		ccid3_pr_debug("%s(%p), s=%u, MSS=%u, "
-			       "R_sample=%uus, X=%u\n", dccp_role(sk),
-			       sk, hctx->ccid3hctx_s,
-			       dccp_sk(sk)->dccps_mss_cache, r_sample,
-			       (unsigned)(hctx->ccid3hctx_x >> 6));
 
 		ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
 
-	} else {
-		/* Update sending rate (step 4 of [RFC 3448, 4.3]) */
-		if (hctx->ccid3hctx_p > 0)
-			hctx->ccid3hctx_x_calc -				tfrc_calc_x(hctx->ccid3hctx_s,
-					    hctx->ccid3hctx_rtt,
-					    hctx->ccid3hctx_p);
-		ccid3_hc_tx_update_x(sk, &now);
-
-		ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
-			       "p=%u, X_calc=%u, X_recv=%u, X=%u\n",
-			       dccp_role(sk),
-			       sk, hctx->ccid3hctx_rtt, r_sample,
-			       hctx->ccid3hctx_s, hctx->ccid3hctx_p,
-			       hctx->ccid3hctx_x_calc,
-			       (unsigned)(hctx->ccid3hctx_x_recv >> 6),
-			       (unsigned)(hctx->ccid3hctx_x >> 6));
+		if (hctx->ccid3hctx_t_rto = 0) {
+			/*
+			 * Initial feedback packet: Larger Initial Windows (4.2)
+			 */
+			hctx->ccid3hctx_x    = rfc3390_initial_rate(sk);
+			hctx->ccid3hctx_t_ld = now;
+
+			ccid3_update_send_interval(hctx);
+
+			goto done_computing_x;
+
+		} else if (hctx->ccid3hctx_p = 0) {
+			/*
+			 * First feedback after nofeedback timer expiry (4.3)
+			 */
+			goto done_computing_x;
+		}
 	}
+	/* perform step (4) of draft rfc3448bis, section 4.3 */
+	if (hctx->ccid3hctx_p > 0)
+		hctx->ccid3hctx_x_calc = tfrc_calc_x(hctx->ccid3hctx_s,
+						     hctx->ccid3hctx_rtt,
+						     hctx->ccid3hctx_p);
+	ccid3_hc_tx_update_x(sk, &now);
+
+done_computing_x:
+	ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, p=%u, X_calc=%u, "
+		       "X_recv=%u, X=%u\n", dccp_role(sk), sk,
+		       hctx->ccid3hctx_rtt, r_sample, hctx->ccid3hctx_s,
+		       hctx->ccid3hctx_p, hctx->ccid3hctx_x_calc,
+		       (unsigned)(hctx->ccid3hctx_x_recv >> 6),
+		       (unsigned)(hctx->ccid3hctx_x >> 6));
 
 	/* unschedule no feedback timer */
 	sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer);

Re: [PATCH 5/5]: Implement rfc3448bis changes to feedback reception

[Ian McDonald]

On 6/12/07, Gerrit Renker <gerrit@erg.abdn.ac.uk> wrote:
> [CCID 3]: Implement rfc3448bis changes to feedback reception
>
> Note: At some point of time this probably needs update wrt rfc3448bis, but see
>       http://www.mail-archive.com/dccp%40ietf.org/msg00426.html
>

I agree that sending rate calculation, particularly after idle still
needs work - but this is at the  IETF level and this helps a lot by
the looks.

> Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>

Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
-- 
Web: http://wand.net.nz/~iam4/
Blog: http://iansblog.jandi.co.nz
WAND Network Research Group

Re: [PATCH 5/5]: Implement rfc3448bis changes to feedback reception

[Gerrit Renker]

Quoting Ian McDonald:
|  > Note: At some point of time this probably needs update wrt rfc3448bis, but see
|  >       http://www.mail-archive.com/dccp%40ietf.org/msg00426.html
|  >
|  
|  I agree that sending rate calculation, particularly after idle still
|  needs work - but this is at the  IETF level and this helps a lot by
|  the looks.
Fully agree with that point and this should be revised as soon as there is some
stabilisation in the specification. Out of my head, the following things more or
less directly depend on this:

 * Oscillation Prevention from RFC 3448 / rfc3448bis
   - affects also `congestion control on LANs' (a point you raised)
   - with the RTTs measured on layer 4, this would be useful
     since when sending `blasts' of packets, the queue buildup is quickly perceivable

 * Faster Restart

 * the CCID 3 definition of idle times - RFC 4342 (6.4) uses the max(0.2 sec, 2*RTT)

 * ... other areas depending on the calculation I may have missed