From: Daniele Lacamera <mlists@danielinux.net>
To: Stephen Hemminger <shemminger@osdl.org>
Cc: "David S. Miller" <davem@davemloft.net>, netdev@oss.sgi.com
Subject: Re: [PATCH] TCP Hybla
Date: Mon, 21 Mar 2005 12:50:07 +0100 [thread overview]
Message-ID: <200503211250.07814.mlists@danielinux.net> (raw)
In-Reply-To: <20050318163123.14969084@dxpl.pdx.osdl.net>
[-- Attachment #1: Type: text/plain, Size: 669 bytes --]
On Saturday 19 March 2005 01:31, Stephen Hemminger wrote:
> Here is a version of TCP Hybla based on the new split out of TCP
> algorithms. It doesn't work right, I probably broke something.
>
> Original code for RTT0 was wrong because HZ=1000 on 2.6, so I changed
> it to be a parameter explicitly in ms.
>
> Don't put it into production system till worked out.
This one is working. I've also made some changes. The cong_avoid checks
for a flightsize not smaller than actual cwnd before giving the
increment. Also, the tcp_reno_cong_avoid is called when not in
TCP_CA_Open, as happens in vegas too.
--
Signed-off by: Daniele Lacamera (root at danielinux.net)
[-- Attachment #2: SH_tcpsplit_hybla.patch --]
[-- Type: text/x-diff, Size: 7610 bytes --]
diff -ruN a/net/ipv4/Kconfig b/net/ipv4/Kconfig
--- a/net/ipv4/Kconfig 2005-03-21 12:17:59.000000000 +0100
+++ b/net/ipv4/Kconfig 2005-03-21 12:15:05.000000000 +0100
@@ -405,6 +405,16 @@
TCP Westwood+ significantly increases fairness wrt TCP Reno in
wired networks and throughput over wireless links.
+config TCP_CONG_HYBLA
+ tristate "TCP-Hybla congestion control algorithm"
+ depends on EXPERIMENTAL
+ default n
+ ---help---
+ TCP-Hybla is a sender-side only change that eliminates penalization of
+ long-RTT, large-bandwidth connections, like when satellite legs are
+ involved, expecially when sharing a common bottleneck with normal
+ terrestrial connections.
+
endmenu
diff -ruN a/net/ipv4/Makefile b/net/ipv4/Makefile
--- a/net/ipv4/Makefile 2005-03-21 12:17:37.000000000 +0100
+++ b/net/ipv4/Makefile 2005-03-21 12:15:05.000000000 +0100
@@ -27,6 +27,7 @@
obj-$(CONFIG_TCP_CONG_VEGAS) += tcp_vegas.o
obj-$(CONFIG_TCP_CONG_BIC) += tcp_bic.o
obj-$(CONFIG_TCP_CONG_WESTWOOD) += tcp_westwood.o
+obj-$(CONFIG_TCP_CONG_HYBLA) += tcp_hybla.o
obj-$(CONFIG_XFRM) += xfrm4_policy.o xfrm4_state.o xfrm4_input.o \
xfrm4_output.o
diff -ruN a/net/ipv4/tcp_hybla.c b/net/ipv4/tcp_hybla.c
--- a/net/ipv4/tcp_hybla.c 1970-01-01 01:00:00.000000000 +0100
+++ b/net/ipv4/tcp_hybla.c 2005-03-21 12:15:06.000000000 +0100
@@ -0,0 +1,207 @@
+/*
+ * TCP HYBLA
+ *
+ * TCP-HYBLA Congestion control algorithm, based on:
+ * C.Caini, R.Firrincieli, "TCP-Hybla: A TCP Enhancement
+ * for Heterogeneous Networks",
+ * International Journal on satellite Communications,
+ * September 2004
+ * Daniele Lacamera
+ * root at danielinux.net
+ */
+
+#include <linux/config.h>
+#include <linux/module.h>
+#include <net/tcp.h>
+
+/* Tcp Hybla structure. */
+struct hybla_ca {
+ u8 hybla_en;
+ u32 snd_cwnd_cents; /* Keeps increment values when it is <1, <<7 */
+ u32 rho; /* Rho parameter, integer part */
+ u32 rho2; /* Rho * Rho, integer part */
+ u32 rho_3ls; /* Rho parameter, <<3 */
+ u32 rho2_7ls; /* Rho^2, <<7 */
+ u32 minrtt; /* Minimum smoothed round trip time value seen */
+};
+
+/* Hybla reference round trip time (default= 1/40 sec = 25 ms),
+ expressed in jiffies */
+static int rtt0 = 25;
+module_param(rtt0, int, 0644);
+MODULE_PARM_DESC(rtt0, "reference rout trip time (ms)");
+
+
+/* This is called to refresh values for hybla parameters */
+static inline void hybla_recalc_param (struct tcp_sock *tp)
+{
+ struct hybla_ca *ca = tcp_ca(tp);
+
+ ca->rho_3ls = max_t(u32, tp->srtt / msecs_to_jiffies(rtt0), 8);
+ ca->rho = ca->rho_3ls >> 3;
+ ca->rho2_7ls = (ca->rho_3ls * ca->rho_3ls) << 1;
+ ca->rho2 = ca->rho2_7ls >>7;
+}
+
+
+static void hybla_start(struct tcp_sock *tp)
+{
+ struct hybla_ca *ca = tcp_ca(tp);
+
+ ca->rho = 0;
+ ca->rho2 = 0;
+ ca->rho_3ls = 0;
+ ca->rho2_7ls = 0;
+ ca->snd_cwnd_cents = 0;
+ ca->hybla_en = 1;
+
+ tp->snd_cwnd = 2;
+ tp->snd_cwnd_clamp=65535;
+}
+
+
+static void hybla_ca_state(struct tcp_sock *tp, u8 ca_state)
+{
+ struct hybla_ca *ca = tcp_ca(tp);
+ if (ca_state == TCP_CA_Open)
+ ca->hybla_en=1;
+ else
+ ca->hybla_en=0;
+}
+
+static inline u32 hybla_fraction(u32 odds)
+{
+ static const u32 fractions[] = {
+ 128, 139, 152, 165, 181, 197, 215, 234,
+ };
+
+ return (odds < ARRAY_SIZE(fractions)) ? fractions[odds] : 128;
+}
+
+/* TCP Hybla main routine.
+ * This is the algorithm behavior:
+ * o Recalc Hybla parameters if min_rtt has changed
+ * o Give cwnd a new value based on the model proposed
+ * o remember increments <1
+ */
+static void hybla_cong_avoid(struct tcp_sock *tp, u32 ack, u32 rtt,
+ u32 in_flight)
+{
+ struct hybla_ca *ca = tcp_ca(tp);
+ u32 increment, odd, rho_fractions;
+ int is_slowstart = 0;
+
+ if(!ca->hybla_en)
+ return tcp_reno_cong_avoid(tp,ack,rtt,in_flight);
+ if (in_flight < tp->snd_cwnd)
+ return;
+
+ if (ca->rho==0)
+ hybla_recalc_param(tp);
+
+ rho_fractions = ca->rho_3ls - (ca->rho << 3);
+
+ if (tp->snd_cwnd < tp->snd_ssthresh) {
+ /*
+ * slow start
+ * INC = 2^RHO - 1
+ * This is done by splitting the rho parameter
+ * into 2 parts: an integer part and a fraction part.
+ * Inrement<<7 is estimated by doing:
+ * [2^(int+fract)]<<7
+ * that is equal to:
+ * (2^int) * [(2^fract) <<7]
+ * 2^int is straightly computed as 1<<int,
+ * while we will use hybla_slowstart_fraction_increment() to
+ * calculate 2^fract in a <<7 value.
+ */
+ is_slowstart = 1;
+ increment = ((1 << ca->rho) * hybla_fraction(rho_fractions))
+ - 128;
+ } else {
+ /*
+ * congestion avoidance
+ * INC = RHO^2 / W
+ * as long as increment is estimated as (rho<<7)/window
+ * it already is <<7 and we can easily count its fractions.
+ */
+ increment = ca->rho2_7ls / tp->snd_cwnd;
+ if (increment < 128)
+ tp->snd_cwnd_cnt++;
+ }
+
+ odd = increment % 128;
+ tp->snd_cwnd += increment >> 7;
+ ca->snd_cwnd_cents += odd;
+
+ /* check when fractions goes >=128 and increase cwnd by 1. */
+ while(ca->snd_cwnd_cents >= 128) {
+ tp->snd_cwnd++;
+ ca->snd_cwnd_cents -= 128;
+ 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);
+
+ tp->snd_cwnd = min_t(u32, tp->snd_cwnd, tp->snd_cwnd_clamp);
+}
+
+/*
+ * Update Values, if necessary, when a new
+ * smoothed RTT Estimation becomes available
+ */
+static void hybla_update_rtt(struct tcp_sock *tp, u32 m)
+{
+ struct hybla_ca *ca = tcp_ca(tp);
+
+ /* This sets rho to the smallest RTT received. */
+ if (tp->srtt) {
+ /* Recalculate rho only if this srtt is the lowest */
+ if (tp->srtt < ca->minrtt){
+ hybla_recalc_param(tp);
+ ca->minrtt = tp->srtt;
+ }
+ } else {
+ /* 1st Rho measurement */
+ hybla_recalc_param(tp);
+
+ /* set minimum rtt as this is the 1st ever seen */
+ ca->minrtt = tp->srtt;
+ tp->snd_cwnd = ca->rho;
+ }
+}
+
+
+
+static struct tcp_ca_type tcp_hybla = {
+ .start = hybla_start,
+ .ssthresh = tcp_reno_ssthresh,
+ .min_cwnd = tcp_reno_cwnd_min,
+ .cong_avoid = hybla_cong_avoid,
+ .rtt_sample = hybla_update_rtt,
+ .set_state = hybla_ca_state,
+
+ .owner = THIS_MODULE,
+ .name = "hybla"
+};
+
+static int __init hybla_init(void)
+{
+ BUG_ON(sizeof(struct hybla_ca) > TCP_CA_PRIV_SIZE);
+ tcp_ca_register(&tcp_hybla);
+ return 0;
+}
+
+static void __exit hybla_exit(void)
+{
+ tcp_ca_unregister(&tcp_hybla);
+}
+
+module_init(hybla_init);
+module_exit(hybla_exit);
+
+MODULE_AUTHOR("Daniele Lacamera");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("TCP Hybla");
prev parent reply other threads:[~2005-03-21 11:50 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2005-03-19 0:31 [PATCH] TCP Hybla Stephen Hemminger
2005-03-19 16:59 ` Daniele Lacamera
2005-03-19 17:19 ` Arnaldo Carvalho de Melo
2005-03-21 11:50 ` Daniele Lacamera [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=200503211250.07814.mlists@danielinux.net \
--to=mlists@danielinux.net \
--cc=davem@davemloft.net \
--cc=netdev@oss.sgi.com \
--cc=shemminger@osdl.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).