dccp ccid-3: Preventing Oscillations

This implements [RFC 3448, 4.5], which performs congestion avoidance behaviour
by reducing the transmit rate as the queueing delay (measured in terms of
long-term RTT) increases.

Oscillation can be turned on/off via a module option (do_osc_prev) and via sysfs
(using mode 0644), the default is off.

Overflow analysis:
------------------
 * oscillation prevention is done after update_x(), so that t_ipi <= 64000;
 * hence the multiplication "t_ipi * sqrt(R_sample)" needs 64 bits;
 * done using u64 for sqrt_sample and explicit typecast of t_ipi;
 * the divisor, R_sqmean, is non-zero because oscillation prevention is first
   called when receiving the second feedback packet, and tfrc_scaled_rtt() > 0.

A detailed discussion of the algorithm (with plots) is on
http://www.erg.abdn.ac.uk/users/gerrit/dccp/notes/ccid3/sender_notes/oscillation_prevention/

The algorithm has negative side effects:
  * when allowing to decrease t_ipi (leads to a large RTT) and
  * when using it during slow-start;
both uses are therefore disabled.

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

1 files changed, 40 insertions, 0 deletions

diff --git a/net/dccp/ccids/ccid3.c b/net/dccp/ccids/ccid3.c
index 7cd76c6c790c..06cfdad84a6a 100644
--- a/net/dccp/ccids/ccid3.c
+++ b/net/dccp/ccids/ccid3.c
@@ -49,6 +49,8 @@ static int ccid3_debug;
 /*
  *      Transmitter Half-Connection Routines
  */
+/* Oscillation Prevention/Reduction: recommended by rfc3448bis, on by default */
+static int do_osc_prev = true;
 
 /*
  * Compute the initial sending rate X_init in the manner of RFC 3390:
@@ -296,6 +298,9 @@ static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
                 hctx->s = ccid3_hc_tx_measure_packet_size(sk, skb->len);
                 ccid3_update_send_interval(hctx);
 
+                /* Seed value for Oscillation Prevention (sec. 4.5) */
+                hctx->r_sqmean = tfrc_scaled_sqrt(hctx->rtt);
+
         } else {
                 delay = ktime_us_delta(hctx->t_nom, now);
                 ccid3_pr_debug("delay=%ld\n", (long)delay);
@@ -400,6 +405,38 @@ done_computing_x:
                                hctx->s, hctx->p, hctx->x_calc,
                                (unsigned)(hctx->x_recv >> 6),
                                (unsigned)(hctx->x >> 6));
+        /*
+         * Oscillation Reduction (RFC 3448, 4.5) - modifying t_ipi according to
+         * RTT changes, multiplying by X/X_inst = sqrt(R_sample)/R_sqmean. This
+         * can be useful if few connections share a link, avoiding that buffer
+         * fill levels (RTT) oscillate as a result of frequent adjustments to X.
+         * A useful presentation with background information is in
+         *    Joerg Widmer, "Equation-Based Congestion Control",
+         *    MSc Thesis, University of Mannheim, Germany, 2000
+         * (sec. 3.6.4), who calls this ISM ("Inter-packet Space Modulation").
+         */
+        if (do_osc_prev) {
+                r_sample = tfrc_scaled_sqrt(r_sample);
+                /*
+                 * The modulation can work in both ways: increase/decrease t_ipi
+                 * according to long-term increases/decreases of the RTT. The
+                 * former is a useful measure, since it works against queue
+                 * build-up. The latter temporarily increases the sending rate,
+                 * so that buffers fill up more quickly. This in turn causes
+                 * the RTT to increase, so that either later reduction becomes
+                 * necessary or the RTT stays at a very high level. Decreasing
+                 * t_ipi is therefore not supported.
+                 * Furthermore, during the initial slow-start phase the RTT
+                 * naturally increases, where using the algorithm would cause
+                 * delays. Hence it is disabled during the initial slow-start.
+                 */
+                if (r_sample > hctx->r_sqmean && hctx->p > 0)
+                        hctx->t_ipi = div_u64((u64)hctx->t_ipi * (u64)r_sample,
+                                              hctx->r_sqmean);
+                hctx->t_ipi = min_t(u32, hctx->t_ipi, TFRC_T_MBI);
+                /* update R_sqmean _after_ computing the modulation factor */
+                hctx->r_sqmean = tfrc_ewma(hctx->r_sqmean, r_sample, 9);
+        }
 
         /* unschedule no feedback timer */
         sk_stop_timer(sk, &hctx->no_feedback_timer);
@@ -749,6 +786,9 @@ static struct ccid_operations ccid3 = {
         .ccid_hc_tx_getsockopt     = ccid3_hc_tx_getsockopt,
 };
 
+module_param(do_osc_prev, bool, 0644);
+MODULE_PARM_DESC(do_osc_prev, "Use Oscillation Prevention (RFC 3448, 4.5)");
+
 #ifdef CONFIG_IP_DCCP_CCID3_DEBUG
 module_param(ccid3_debug, bool, 0644);
 MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");