On Mon, 15 Mar 2004 13:00:43 -0800 "David S. Miller" <davem@redhat.com> wrote: >On Mon, 15 Mar 2004 16:42:20 +0100 >"Angelo Dell'Aera" <buffer@antifork.org> wrote: > >> this is the 2.4 version of Stephen's patch which removes >> bw_sample. I did even few small changes to comments which >> were simply too old and didn't reflect in any way the >> current implementation. It applies against 2.4.26-pre3. > >Applied, can you make the comment fixups (on top of Stephen's >patch) for 2.6.x too? We should keep these two things in sync >as far as is reasonable. That's what you asked for me. It applies against 2.6.4 with Stephen's patch still applied. Regards. -- Angelo Dell'Aera 'buffer' Antifork Research, Inc. http://buffer.antifork.org diff -Naur linux-2.6.4-old/net/ipv4/tcp_input.c linux-2.6.4/net/ipv4/tcp_input.c --- linux-2.6.4-old/net/ipv4/tcp_input.c 2004-03-15 23:30:13.000000000 +0100 +++ linux-2.6.4/net/ipv4/tcp_input.c 2004-03-15 23:36:44.000000000 +0100 @@ -2051,16 +2051,21 @@ */ tp->frto_counter = (tp->frto_counter + 1) % 3; } + /* - * TCP Westwood - * Functions needed for estimating bandwidth. + * TCP Westwood+ */ /* - * This function initializes fields used in TCP Westwood. - * We can't get no information about RTT at this time so - * we are forced to set it to 0. + * @init_westwood + * This function initializes fields used in TCP Westwood+. We can't + * get no information about RTTmin at this time so we simply set it to + * TCP_WESTWOOD_INIT_RTT. This value was chosen to be too conservative + * since in this way we're sure it will be updated in a consistent + * way as soon as possible. It will reasonably happen within the first + * RTT period of the connection lifetime. */ + static void init_westwood(struct sock *sk) { struct tcp_opt *tp = tcp_sk(sk); @@ -2079,6 +2084,7 @@ * @westwood_do_filter * Low-pass filter. Implemented using constant coeffients. */ + static inline __u32 westwood_do_filter(__u32 a, __u32 b) { return (((7 * a) + b) >> 3); @@ -2096,10 +2102,12 @@ tp->westwood.bw_ns_est); } -/* @westwood_update_rttmin +/* + * @westwood_update_rttmin * It is used to update RTTmin. In this case we MUST NOT use * WESTWOOD_RTT_MIN minimum bound since we could be on a LAN! */ + static inline __u32 westwood_update_rttmin(struct sock *sk) { struct tcp_opt *tp = tcp_sk(sk); @@ -2116,9 +2124,9 @@ /* * @westwood_acked - * Evaluate increases for dk. It requires no lock since when it is - * called lock should already be held. Be careful about it! + * Evaluate increases for dk. */ + static inline __u32 westwood_acked(struct sock *sk) { struct tcp_opt *tp = tcp_sk(sk); @@ -2134,6 +2142,7 @@ * It returns 0 if we are still evaluating samples in the same RTT * window, 1 if the sample has to be considered in the next window. */ + static int westwood_new_window(struct sock *sk) { struct tcp_opt *tp = tcp_sk(sk); @@ -2162,11 +2171,9 @@ /* * @westwood_update_window * It updates RTT evaluation window if it is the right moment to do - * it. If so it calls filter for evaluating bandwidth. Be careful - * about __westwood_update_window() since it is called without - * any form of lock. It should be used only for internal purposes. - * Call westwood_update_window() instead. + * it. If so it calls filter for evaluating bandwidth. */ + static void __westwood_update_window(struct sock *sk, __u32 now) { struct tcp_opt *tp = tcp_sk(sk); @@ -2190,11 +2197,12 @@ } /* - * @__westwood_fast_bw + * @__tcp_westwood_fast_bw * It is called when we are in fast path. In particular it is called when * header prediction is successfull. In such case infact update is * straight forward and doesn't need any particular care. */ + void __tcp_westwood_fast_bw(struct sock *sk, struct sk_buff *skb) { struct tcp_opt *tp = tcp_sk(sk); @@ -2208,7 +2216,7 @@ /* - * @tcp_westwood_dupack_update + * @westwood_dupack_update * It updates accounted and cumul_ack when receiving a dupack. */ @@ -2242,6 +2250,7 @@ * This function evaluates cumul_ack for evaluating dk in case of * delayed or partial acks. */ + static __u32 westwood_acked_count(struct sock *sk) { struct tcp_opt *tp = tcp_sk(sk); @@ -2269,11 +2278,12 @@ /* - * @__westwood_slow_bw + * @__tcp_westwood_slow_bw * It is called when something is going wrong..even if there could * be no problems! Infact a simple delayed packet may trigger a * dupack. But we need to be careful in such case. */ + void __tcp_westwood_slow_bw(struct sock *sk, struct sk_buff *skb) { struct tcp_opt *tp = tcp_sk(sk); - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
