tcp: Fix tcp_hybla zero congestion window growth with small rho and large cwnd.
authorDaniele Lacamera <root@danielinux.net>
Tue, 7 Oct 2008 22:58:17 +0000 (15:58 -0700)
committerDavid S. Miller <davem@davemloft.net>
Tue, 7 Oct 2008 22:58:17 +0000 (15:58 -0700)
Because of rounding, in certain conditions, i.e. when in congestion
avoidance state rho is smaller than 1/128 of the current cwnd, TCP
Hybla congestion control starves and the cwnd is kept constant
forever.

This patch forces an increment by one segment after #send_cwnd calls
without increments(newreno behavior).

Signed-off-by: Daniele Lacamera <root@danielinux.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/tcp_hybla.c

index bfcbd148a89d5daec4bf09b169010e6690c6f6c8..c209e054a634ed5b07f9b6d8374006887a2c2085 100644 (file)
@@ -150,7 +150,11 @@ static void hybla_cong_avoid(struct sock *sk, u32 ack, u32 in_flight)
                ca->snd_cwnd_cents -= 128;
                tp->snd_cwnd_cnt = 0;
        }
-
+       /* check when cwnd has not been incremented for a while */
+       if (increment == 0 && odd == 0 && tp->snd_cwnd_cnt >= tp->snd_cwnd) {
+               tp->snd_cwnd++;
+               tp->snd_cwnd_cnt = 0;
+       }
        /* clamp down slowstart cwnd to ssthresh value. */
        if (is_slowstart)
                tp->snd_cwnd = min(tp->snd_cwnd, tp->snd_ssthresh);