Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751106AbWEAKFy (ORCPT ); Mon, 1 May 2006 06:05:54 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1751120AbWEAKFy (ORCPT ); Mon, 1 May 2006 06:05:54 -0400 Received: from pproxy.gmail.com ([64.233.166.181]:14084 "EHLO pproxy.gmail.com") by vger.kernel.org with ESMTP id S1751078AbWEAKFw convert rfc822-to-8bit (ORCPT ); Mon, 1 May 2006 06:05:52 -0400 DomainKey-Signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:cc:mime-version:content-type:content-transfer-encoding:content-disposition; b=XiP11S2k6hLIS4QH7uZ5pEsSUQDMZGGRb54lNEpP4zytNt04kIZB5gZUIotxHaOuK6QjGAMX9lOT96jldrUtftoE/TDqwl8PquSCXvkU5NOsieKGsHE/R8PtcRN/Ievq5L2j9yXEZufyE2Hooh3UPZ+3BU5GaW84iFJy0ZIwy6Q= Message-ID: <3feffd230605010305o6e3b1511of1f75b17f2797e66@mail.gmail.com> Date: Mon, 1 May 2006 18:05:52 +0800 From: "Wong Edison" To: netdev@vger.kernel.org Subject: [PATCH 001/100] TCP congestion module: add TCP-LP supporting for 2.6.16 Cc: linux-kernel@vger.kernel.org MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII; format=flowed Content-Transfer-Encoding: 7BIT Content-Disposition: inline Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 12416 Lines: 420 TCP Low Priority is a distributed algorithm whose goal is to utilize only the excess network bandwidth as compared to the ``fair share`` of bandwidth as targeted by TCP. Available from: http://www.ece.rice.edu/~akuzma/Doc/akuzma/TCP-LP.pdf See http://www-ece.rice.edu/networks/TCP-LP/ for their implementation. Our group take the following changes from the original TCP-LP implementation: o We use newReno in most core CA handling. Only add some checking within cong_avoid. o Error correcting in remote HZ, therefore remote HZ will be keeped on checking and updating. o Handling calculation of One-Way-Delay (OWD) within rtt_sample, sicne OWD have a similar meaning as RTT. Also correct the buggy formular. o Handle reaction for Early Congestion Indication (ECI) within pkts_acked, as mentioned within pseudo code. o OWD is handled in relative format, where local time stamp will in tcp_time_stamp format. Port from 2.4.19 to 2.6.16 as module by: Wong Hoi Sing Edison Hung Hing Lun Signed-off-by: Wong Hoi Sing Edison diff -urN linux-2.6.16.1/net/ipv4/Kconfig linux/net/ipv4/Kconfig --- linux-2.6.16.1/net/ipv4/Kconfig 2006-03-28 14:49:02.000000000 +0800 +++ linux/net/ipv4/Kconfig 2006-04-19 02:40:27.000000000 +0800 @@ -531,6 +531,27 @@ properties, though is known to have fairness issues. See http://www-lce.eng.cam.ac.uk/~ctk21/scalable/ +config TCP_CONG_LP + tristate "TCP Low Priority" + depends on EXPERIMENTAL + default n + ---help--- + TCP Low Priority (TCP-LP), a distributed algorithm whose goal is + to utiliza only the excess network bandwidth as compared to the + ``fair share`` of bandwidth as targeted by TCP. + See http://www-ece.rice.edu/networks/TCP-LP/ + +config TCP_CONG_LP_DEBUG + bool "TCP-LP Debug" + depends on TCP_CONG_LP + default n + ---help--- + Turn on/off the debug message for TCP-LP. The debug message will + print to default kernel debug log file, e.g. /var/log/debug as + default. You can use dmesg to obtain the log too. + + If unsure, say N. + endmenu config TCP_CONG_BIC diff -urN linux-2.6.16.1/net/ipv4/Makefile linux/net/ipv4/Makefile --- linux-2.6.16.1/net/ipv4/Makefile 2006-03-28 14:49:02.000000000 +0800 +++ linux/net/ipv4/Makefile 2006-04-19 02:40:27.000000000 +0800 @@ -41,6 +41,7 @@ obj-$(CONFIG_TCP_CONG_HTCP) += tcp_htcp.o obj-$(CONFIG_TCP_CONG_VEGAS) += tcp_vegas.o obj-$(CONFIG_TCP_CONG_SCALABLE) += tcp_scalable.o +obj-$(CONFIG_TCP_CONG_LP) += tcp_lp.o obj-$(CONFIG_XFRM) += xfrm4_policy.o xfrm4_state.o xfrm4_input.o \ xfrm4_output.o diff -urN linux-2.6.16.1/net/ipv4/tcp_lp.c linux/net/ipv4/tcp_lp.c --- linux-2.6.16.1/net/ipv4/tcp_lp.c 1970-01-01 08:00:00.000000000 +0800 +++ linux/net/ipv4/tcp_lp.c 2006-04-25 15:54:54.000000000 +0800 @@ -0,0 +1,343 @@ +/* + * TCP Low Priority (TCP-LP) + * + * TCP Low Priority is a distributed algorithm whose goal is to utilize only + * the excess network bandwidth as compared to the ``fair share`` of + * bandwidth as targeted by TCP. Available from: + * http://www.ece.rice.edu/~akuzma/Doc/akuzma/TCP-LP.pdf + * + * Original Author: + * Aleksandar Kuzmanovic + * + * See http://www-ece.rice.edu/networks/TCP-LP/ for their implementation. + * As of 2.6.13, Linux supports pluggable congestion control algorithms. + * Due to the limitation of the API, we take the following changes from + * the original TCP-LP implementation: + * o We use newReno in most core CA handling. Only add some checking + * within cong_avoid. + * o Error correcting in remote HZ, therefore remote HZ will be keeped + * on checking and updating. + * o Handling calculation of One-Way-Delay (OWD) within rtt_sample, sicne + * OWD have a similar meaning as RTT. Also correct the buggy formular. + * o Handle reaction for Early Congestion Indication (ECI) within + * pkts_acked, as mentioned within pseudo code. + * o OWD is handled in relative format, where local time stamp will in + * tcp_time_stamp format. + * + * Port from 2.4.19 to 2.6.16 as module by: + * Wong Hoi Sing Edison + * Hung Hing Lun + * + * Version: $Id: tcp_lp.c,v 1.20 2006-04-22 06:34:20 hswong3i Exp $ + */ + +#include +#include +#include + +#ifndef CONFIG_TCP_CONG_LP_DEBUG +#define CONFIG_TCP_CONG_LP_DEBUG 0 +#endif + +/* resolution of owd */ +#define LP_RESOL 1000 + +/** + * enum tcp_lp_state + * @LP_VALID_RHZ: is remote HZ valid? + * @LP_VALID_OWD: is OWD valid? + * @LP_WITHIN_THR: are we within threshold? + * @LP_WITHIN_INF: are we within inference? + * + * TCP-LP's state flags. + * We create this set of state flag mainly for debugging. + */ +enum tcp_lp_state { + LP_VALID_RHZ = (1 << 0), + LP_VALID_OWD = (1 << 1), + LP_WITHIN_THR = (1 << 3), + LP_WITHIN_INF = (1 << 4), +}; + +/** + * struct lp + * @flag: TCP-LP state flag + * @sowd: smoothed OWD << 3 + * @owd_min: min OWD + * @owd_max: max OWD + * @owd_max_rsv: resrved max owd + * @RHZ: estimated remote HZ + * @remote_ref_time: remote reference time + * @local_ref_time: local reference time + * @last_drop: time for last active drop + * @inference: current inference + * + * TCP-LP's private struct. + * We get the idea from original TCP-LP implementation where only left those we + * found are really useful. + */ +struct lp { + u32 flag; + u32 sowd; + u32 owd_min; + u32 owd_max; + u32 owd_max_rsv; + u32 RHZ; + u32 remote_ref_time; + u32 local_ref_time; + u32 last_drop; + u32 inference; +}; + +/** + * tcp_lp_init + * + * Init all required variables. + * Clone the handling from Vegas module implementation. + */ +static void tcp_lp_init(struct sock *sk) +{ + struct lp *lp = inet_csk_ca(sk); + + lp->flag = 0; + lp->sowd = 0; + lp->owd_min = 0xffffffff; + lp->owd_max = 0; + lp->owd_max_rsv = 0; + lp->RHZ = 0; + lp->remote_ref_time = 0; + lp->local_ref_time = 0; + lp->last_drop = 0; + lp->inference = 0; +} + +/** + * tcp_lp_cong_avoid + * + * Implementation of cong_avoid. + * Will only call newReno CA when away from inference. + * From TCP-LP's paper, this will be handled in additive increasement. + */ +static void tcp_lp_cong_avoid(struct sock *sk, u32 ack, u32 rtt, u32 in_flight, + int flag) +{ + struct lp *lp = inet_csk_ca(sk); + + if (!(lp->flag & LP_WITHIN_INF)) + tcp_reno_cong_avoid(sk, ack, rtt, in_flight, flag); +} + +/** + * tcp_lp_remote_hz_estimator + * + * Estimate remote HZ. + * We keep on updating the estimated value, where original TCP-LP + * implementation only guest it for once and use forever. + */ +static inline u32 tcp_lp_remote_hz_estimator(struct sock *sk) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct lp *lp = inet_csk_ca(sk); + s64 rhz = lp->RHZ << 6; /* remote HZ << 6 */ + s64 m = 0; + + /* not yet record reference time + * go away!! record it before come back!! */ + if (lp->remote_ref_time == 0 || lp->local_ref_time == 0) + goto out; + + /* we can't calc remote HZ with no different!! */ + if (tp->rx_opt.rcv_tsval == lp->remote_ref_time + || tp->rx_opt.rcv_tsecr == lp->local_ref_time) + goto out; + + m = HZ * (tp->rx_opt.rcv_tsval - + lp->remote_ref_time) / (tp->rx_opt.rcv_tsecr - + lp->local_ref_time); + if (m < 0) + m = -m; + + if (rhz != 0) { + m -= (rhz >> 6); /* m is now error in remote HZ est */ + rhz += m; /* 63/64 old + 1/64 new */ + } else + rhz = m << 6; + + /* record time for successful remote HZ calc */ + lp->flag |= LP_VALID_RHZ; + + out: + /* record reference time stamp */ + lp->remote_ref_time = tp->rx_opt.rcv_tsval; + lp->local_ref_time = tp->rx_opt.rcv_tsecr; + + return rhz >> 6; +} + +/** + * tcp_lp_owd_calculator + * + * Calculate one way delay (in relative format). + * Original implement OWD as minus of remote time difference to local time + * difference directly. As this time difference just simply equal to RTT, when + * the network status is stable, remote RTT will equal to local RTT, and result + * OWD into zero. + * It seems to be a bug and so we fixed it. + */ +static inline u32 tcp_lp_owd_calculator(struct sock *sk) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct lp *lp = inet_csk_ca(sk); + s64 owd = 0; + + lp->RHZ = tcp_lp_remote_hz_estimator(sk); + + if (lp->flag & LP_VALID_RHZ) { + owd = + tp->rx_opt.rcv_tsval * (LP_RESOL / lp->RHZ) - + tp->rx_opt.rcv_tsecr * (LP_RESOL / HZ); + if (owd < 0) + owd = -owd; + } + + if (owd > 0) + lp->flag |= LP_VALID_OWD; + else + lp->flag &= ~LP_VALID_OWD; + + return owd; +} + +/** + * tcp_lp_rtt_sample + * + * Implementation or rtt_sample. + * Will take the following action, + * 1. calc OWD, + * 2. record the min/max OWD, + * 3. calc smoothed OWD (SOWD). + * Most ideas come from the original TCP-LP implementation. + */ +static void tcp_lp_rtt_sample(struct sock *sk, u32 usrtt) +{ + struct lp *lp = inet_csk_ca(sk); + s64 mowd = tcp_lp_owd_calculator(sk); + + /* sorry that we don't have valid data */ + if (!(lp->flag & LP_VALID_RHZ) || !(lp->flag & LP_VALID_OWD)) + return; + + /* record the next min owd */ + if (mowd < lp->owd_min) + lp->owd_min = mowd; + + /* always forget the max of the max + * we just set owd_max as one below it */ + if (mowd > lp->owd_max) { + if (mowd > lp->owd_max_rsv) { + if (lp->owd_max_rsv == 0) + lp->owd_max = mowd; + else + lp->owd_max = lp->owd_max_rsv; + lp->owd_max_rsv = mowd; + } else + lp->owd_max = mowd; + } + + /* calc for smoothed owd */ + if (lp->sowd != 0) { + mowd -= (lp->sowd >> 3); /* m is now error in owd est */ + lp->sowd += mowd; /* owd = 7/8 owd + 1/8 new */ + } else + lp->sowd = mowd << 3; /* take the measured time be owd */ +} + +/** + * tcp_lp_pkts_acked + * + * Implementation of pkts_acked. + * Deal with active drop under Early Congestion Indication. + * Only drop to half and 1 will be handle, because we hope to use back + * newReno in increase case. + * We work it out by following the idea from TCP-LP's paper directly + */ +static void tcp_lp_pkts_acked(struct sock *sk, u32 num_acked) +{ + struct tcp_sock *tp = tcp_sk(sk); + struct lp *lp = inet_csk_ca(sk); + + /* calc inference */ + if (tcp_time_stamp > tp->rx_opt.rcv_tsecr) + lp->inference = 3 * (tcp_time_stamp - tp->rx_opt.rcv_tsecr); + + /* test if within inference */ + if (lp->last_drop && (tcp_time_stamp - lp->last_drop < lp->inference)) + lp->flag |= LP_WITHIN_INF; + else + lp->flag &= ~LP_WITHIN_INF; + + /* test if within threshold */ + if (lp->sowd >> 3 < + lp->owd_min + 15 * (lp->owd_max - lp->owd_min) / 100) + lp->flag |= LP_WITHIN_THR; + else + lp->flag &= ~LP_WITHIN_THR; + +#if CONFIG_TCP_CONG_LP_DEBUG == 1 + printk(KERN_DEBUG "TCP-LP: %05o|%5u|%5u|%15u|%15u|%15u\n", lp->flag, + tp->snd_cwnd, lp->RHZ, lp->owd_min, lp->owd_max, lp->sowd >> 3); +#endif + + if (lp->flag & LP_WITHIN_THR) + return; + + /* FIXME: try to reset owd_min and owd_max here + * so decrease the chance the min/max is no longer suitable + * and will usually within threshold when whithin inference */ + lp->owd_min = (lp->sowd >> 3); + lp->owd_max = (lp->sowd >> 2); + lp->owd_max_rsv = (lp->sowd >> 2); + + /* happened within inference + * drop snd_cwnd into 1 */ + if (lp->flag & LP_WITHIN_INF) + tp->snd_cwnd = 1U; + + /* happened after inference + * cut snd_cwnd into half */ + else + tp->snd_cwnd = max(tp->snd_cwnd >> 1U, 1U); + + /* record this drop time */ + lp->last_drop = tcp_time_stamp; +} + +static struct tcp_congestion_ops tcp_lp = { + .init = tcp_lp_init, + .ssthresh = tcp_reno_ssthresh, + .cong_avoid = tcp_lp_cong_avoid, + .min_cwnd = tcp_reno_min_cwnd, + .rtt_sample = tcp_lp_rtt_sample, + .pkts_acked = tcp_lp_pkts_acked, + + .owner = THIS_MODULE, + .name = "lp" +}; + +static int __init lp_register(void) +{ + BUG_ON(sizeof(struct lp) > ICSK_CA_PRIV_SIZE); + return tcp_register_congestion_control(&tcp_lp); +} + +static void __exit lp_unregister(void) +{ + tcp_unregister_congestion_control(&tcp_lp); +} + +module_init(lp_register); +module_exit(lp_unregister); + +MODULE_AUTHOR("Wong Hoi Sing Edison, Hung Hing Lun"); +MODULE_LICENSE("GPL"); +MODULE_DESCRIPTION("TCP Low Priority"); - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/