diff options
author | Gerrit Renker <gerrit@erg.abdn.ac.uk> | 2006-11-28 19:51:42 -0200 |
---|---|---|
committer | David S. Miller <davem@sunset.davemloft.net> | 2006-12-02 21:31:01 -0800 |
commit | a79ef76f4d8424324c2f108824a7398571193f43 (patch) | |
tree | a80e84d19c867e1432fc05ae50199487ce2ea2b2 /net/dccp | |
parent | 841bac1d607d8bf2e068e4b24393fb77372814e3 (diff) | |
download | kernel-crypto-a79ef76f4d8424324c2f108824a7398571193f43.tar.gz kernel-crypto-a79ef76f4d8424324c2f108824a7398571193f43.tar.xz kernel-crypto-a79ef76f4d8424324c2f108824a7398571193f43.zip |
[DCCP] ccid3: Larger initial windows
This implements the larger-initial-windows feature for CCID 3, as described in
section 5 of RFC 4342. When the first feedback packet arrives, the sender can
send up to 2..4 packets per RTT, instead of just one.
The patch further
* reduces the number of timestamping calls by passing the timestamp value
(which is computed in one of the calling functions anyway) as argument
* renames one constant with a very long name into one which is shorter and
resembles the one in RFC 3448 (t_mbi)
* simplifies some of the min_t/max_t cases where both `x', `y' have the same
type
Commiter note: renamed TFRC_t_mbi to TFRC_T_MBI, to follow Linux coding style.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Acked-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
Diffstat (limited to 'net/dccp')
-rw-r--r-- | net/dccp/ccids/ccid3.c | 66 | ||||
-rw-r--r-- | net/dccp/ccids/ccid3.h | 4 |
2 files changed, 37 insertions, 33 deletions
diff --git a/net/dccp/ccids/ccid3.c b/net/dccp/ccids/ccid3.c index 05513f3df65..aa5440ee20a 100644 --- a/net/dccp/ccids/ccid3.c +++ b/net/dccp/ccids/ccid3.c @@ -128,7 +128,8 @@ static inline void ccid3_update_send_time(struct ccid3_hc_tx_sock *hctx) * X = max(min(2 * X, 2 * X_recv), s / R); * tld = now; */ -static void ccid3_hc_tx_update_x(struct sock *sk) +static void ccid3_hc_tx_update_x(struct sock *sk, struct timeval *now) + { struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk); const __u32 old_x = hctx->ccid3hctx_x; @@ -138,23 +139,20 @@ static void ccid3_hc_tx_update_x(struct sock *sk) hctx->ccid3hctx_x_calc = tfrc_calc_x(hctx->ccid3hctx_s, hctx->ccid3hctx_rtt, hctx->ccid3hctx_p); - hctx->ccid3hctx_x = max_t(u32, min_t(u32, hctx->ccid3hctx_x_calc, - 2 * hctx->ccid3hctx_x_recv), - (hctx->ccid3hctx_s / - TFRC_MAX_BACK_OFF_TIME)); - } else { - struct timeval now; + hctx->ccid3hctx_x = max_t(u32, min(hctx->ccid3hctx_x_calc, + hctx->ccid3hctx_x_recv * 2), + hctx->ccid3hctx_s / TFRC_T_MBI); + + } else if (timeval_delta(now, &hctx->ccid3hctx_t_ld) >= + hctx->ccid3hctx_rtt) { + hctx->ccid3hctx_x = max(min(hctx->ccid3hctx_x_recv, + hctx->ccid3hctx_x ) * 2, + usecs_div(hctx->ccid3hctx_s, + hctx->ccid3hctx_rtt) ); + hctx->ccid3hctx_t_ld = *now; + } else + ccid3_pr_debug("Not changing X\n"); - dccp_timestamp(sk, &now); - if (timeval_delta(&now, &hctx->ccid3hctx_t_ld) >= - hctx->ccid3hctx_rtt) { - hctx->ccid3hctx_x = max_t(u32, min_t(u32, hctx->ccid3hctx_x_recv, - hctx->ccid3hctx_x) * 2, - usecs_div(hctx->ccid3hctx_s, - hctx->ccid3hctx_rtt)); - hctx->ccid3hctx_t_ld = now; - } - } if (hctx->ccid3hctx_x != old_x) ccid3_update_send_time(hctx); } @@ -196,12 +194,9 @@ static void ccid3_hc_tx_no_feedback_timer(unsigned long data) switch (hctx->ccid3hctx_state) { case TFRC_SSTATE_NO_FBACK: - /* Halve send rate */ - hctx->ccid3hctx_x /= 2; - if (hctx->ccid3hctx_x < (hctx->ccid3hctx_s / - TFRC_MAX_BACK_OFF_TIME)) - hctx->ccid3hctx_x = (hctx->ccid3hctx_s / - TFRC_MAX_BACK_OFF_TIME); + /* RFC 3448, 4.4: Halve send rate directly */ + hctx->ccid3hctx_x = min_t(u32, hctx->ccid3hctx_x / 2, + hctx->ccid3hctx_s / TFRC_T_MBI); ccid3_pr_debug("%s, sk=%p, state=%s, updated tx rate to %d " "bytes/s\n", @@ -221,6 +216,8 @@ static void ccid3_hc_tx_no_feedback_timer(unsigned long data) if (!hctx->ccid3hctx_idle || (hctx->ccid3hctx_x_recv >= 4 * usecs_div(hctx->ccid3hctx_s, hctx->ccid3hctx_rtt))) { + struct timeval now; + ccid3_pr_debug("%s, sk=%p, state=%s, not idle\n", dccp_role(sk), sk, ccid3_tx_state_name(hctx->ccid3hctx_state)); @@ -238,12 +235,13 @@ static void ccid3_hc_tx_no_feedback_timer(unsigned long data) if (hctx->ccid3hctx_p < TFRC_SMALLEST_P || hctx->ccid3hctx_x_calc > 2 * hctx->ccid3hctx_x_recv) hctx->ccid3hctx_x_recv = max_t(u32, hctx->ccid3hctx_x_recv / 2, - hctx->ccid3hctx_s / (2 * TFRC_MAX_BACK_OFF_TIME)); + hctx->ccid3hctx_s / (2 * TFRC_T_MBI)); else hctx->ccid3hctx_x_recv = hctx->ccid3hctx_x_calc / 4; /* Update sending rate */ - ccid3_hc_tx_update_x(sk); + dccp_timestamp(sk, &now); + ccid3_hc_tx_update_x(sk, &now); } /* * Schedule no feedback timer to expire in @@ -473,11 +471,21 @@ static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb) * q is a constant, RFC 3448 recomments 0.9 */ if (hctx->ccid3hctx_state == TFRC_SSTATE_NO_FBACK) { + /* Use Larger Initial Windows [RFC 4342, sec. 5] + * We deviate in that we use `s' instead of `MSS'. */ + u16 w_init = max( 4 * hctx->ccid3hctx_s, + max(2 * hctx->ccid3hctx_s, 4380)); + hctx->ccid3hctx_rtt = r_sample; + hctx->ccid3hctx_x = usecs_div(w_init, r_sample); + hctx->ccid3hctx_t_ld = now; + + ccid3_update_send_time(hctx); ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK); - hctx->ccid3hctx_rtt = r_sample; - } else + } else { hctx->ccid3hctx_rtt = (hctx->ccid3hctx_rtt * 9) / 10 + r_sample / 10; + ccid3_hc_tx_update_x(sk, &now); + } ccid3_pr_debug("%s, sk=%p, New RTT estimate=%uus, " "r_sample=%us\n", dccp_role(sk), sk, @@ -502,9 +510,6 @@ static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb) /* unschedule no feedback timer */ sk_stop_timer(sk, &hctx->ccid3hctx_no_feedback_timer); - /* Update sending rate (and likely t_ipi, t_nom, and delta) */ - ccid3_hc_tx_update_x(sk); - /* remove all packets older than the one acked from history */ dccp_tx_hist_purge_older(ccid3_tx_hist, &hctx->ccid3hctx_hist, packet); @@ -514,7 +519,6 @@ static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb) */ sk->sk_write_space(sk); - /* Update timeout interval. We use the alternative variant of * [RFC 3448, 3.1] which sets the upper bound of t_rto to one * second, as it is suggested for TCP (see RFC 2988, 2.4). */ diff --git a/net/dccp/ccids/ccid3.h b/net/dccp/ccids/ccid3.h index dbb884426df..27cb20ae1da 100644 --- a/net/dccp/ccids/ccid3.h +++ b/net/dccp/ccids/ccid3.h @@ -48,8 +48,8 @@ /* In usecs - half the scheduling granularity as per RFC3448 4.6 */ #define TFRC_OPSYS_HALF_TIME_GRAN (USEC_PER_SEC / (2 * HZ)) -/* In seconds */ -#define TFRC_MAX_BACK_OFF_TIME 64 +/* Parameter t_mbi from [RFC 3448, 4.3]: backoff interval in seconds */ +#define TFRC_T_MBI 64 #define TFRC_SMALLEST_P 40 |