[TCP] slow start: Make comments and code logic clearer.
authorStephen Hemminger <shemminger@linux-foundation.org>
Thu, 17 May 2007 07:04:18 +0000 (00:04 -0700)
committerDavid S. Miller <davem@sunset.davemloft.net>
Thu, 17 May 2007 21:20:31 +0000 (14:20 -0700)
Add more comments to describe our version of tcp_slow_start().

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/tcp_cong.c

index 86b26539e54b8a8572919acd9c3880fe232ec89a..1260e52ad77286cc4d3022fb8afe1bbdf422cccb 100644 (file)
@@ -276,30 +276,34 @@ int tcp_set_congestion_control(struct sock *sk, const char *name)
 
 
 /*
- * Slow start (exponential increase) with
- * RFC3742 Limited Slow Start (fast linear increase) support.
+ * Slow start is used when congestion window is less than slow start
+ * threshold. This version implements the basic RFC2581 version
+ * and optionally supports:
+ *     RFC3742 Limited Slow Start        - growth limited to max_ssthresh
+ *     RFC3465 Appropriate Byte Counting - growth limited by bytes acknowledged
  */
 void tcp_slow_start(struct tcp_sock *tp)
 {
-       int cnt = 0;
-
-       if (sysctl_tcp_abc) {
-               /* RFC3465: Slow Start
-                * TCP sender SHOULD increase cwnd by the number of
-                * previously unacknowledged bytes ACKed by each incoming
-                * acknowledgment, provided the increase is not more than L
-                */
-               if (tp->bytes_acked < tp->mss_cache)
-                       return;
-       }
+       int cnt; /* increase in packets */
+
+       /* RFC3465: ABC Slow start
+        * Increase only after a full MSS of bytes is acked
+        *
+        * TCP sender SHOULD increase cwnd by the number of
+        * previously unacknowledged bytes ACKed by each incoming
+        * acknowledgment, provided the increase is not more than L
+        */
+       if (sysctl_tcp_abc && tp->bytes_acked < tp->mss_cache)
+               return;
 
-       if (sysctl_tcp_max_ssthresh > 0 &&
-           tp->snd_cwnd > sysctl_tcp_max_ssthresh)
-               cnt += sysctl_tcp_max_ssthresh>>1;
+       if (sysctl_tcp_max_ssthresh > 0 && tp->snd_cwnd > sysctl_tcp_max_ssthresh)
+               cnt = sysctl_tcp_max_ssthresh >> 1;     /* limited slow start */
        else
-               cnt += tp->snd_cwnd;
+               cnt = tp->snd_cwnd;                     /* exponential increase */
 
-       /* RFC3465: We MAY increase by 2 if discovered delayed ack */
+       /* RFC3465: ABC
+        * We MAY increase by 2 if discovered delayed ack
+        */
        if (sysctl_tcp_abc > 1 && tp->bytes_acked >= 2*tp->mss_cache)
                cnt <<= 1;
        tp->bytes_acked = 0;