| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Thu, 08 Dec 2011 17:06:03 +0100
From: Eric Dumazet <eric.dumazet@...il.com>
To: Dave Taht <dave.taht@...il.com>, David Miller <davem@...emloft.net>
Cc: bloat <bloat@...ts.bufferbloat.net>,
bloat-devel <bloat-devel@...ts.bufferbloat.net>,
netdev@...r.kernel.org,
linux-wireless <linux-wireless@...r.kernel.org>,
Hagen Paul Pfeifer <hagen@...u.net>
Subject: [PATCH net-next] sch_red: Adaptative RED AQM
Adaptative RED AQM for linux, based on paper from Sally FLoyd,
Ramakrishna Gummadi, and Scott Shenker, August 2001 :
http://icir.org/floyd/papers/adaptiveRed.pdf
Goal of Adaptative RED is to make max_p a dynamic value between 1% and
50% to reach the target average queue : (max_th - min_th) / 2
Every 500 ms:
if (avg > target and max_p <= 0.5)
increase max_p : max_p += alpha;
else if (avg < target and max_p >= 0.01)
decrease max_p : max_p *= beta;
target :[min_th + 0.4*(min_th - max_th),
min_th + 0.6*(min_th - max_th)].
alpha : min(0.01, max_p / 4)
beta : 0.9
max_P is a Q0.32 fixed point number (unsigned, with 32 bits mantissa)
Changes against our RED implementation are :
max_p is no longer a negative power of two (1/(2^Plog)), but a Q0.32
fixed point number, to allow full range described in Adatative paper.
To deliver a random number, we now use a reciprocal divide (thats really
a multiply), but this operation is done once per marked/droped packet
when in RED_BETWEEN_TRESH window, so added cost (compared to previous
AND operation) is near zero.
dump operation gives current max_p value in a new TCA_RED_MAX_P
attribute.
Example on a 10Mbit link :
tc qdisc add dev $DEV parent 1:1 handle 10: est 1sec 8sec red \
limit 400000 min 30000 max 90000 avpkt 1000 \
burst 55 ecn adaptative bandwidth 10Mbit
# tc -s -d qdisc show dev eth3
...
qdisc red 10: parent 1:1 limit 400000b min 30000b max 90000b ecn
adaptative ewma 5 max_p=0.113335 Scell_log 15
Sent 50414282 bytes 34504 pkt (dropped 35, overlimits 1392 requeues 0)
rate 9749Kbit 831pps backlog 72056b 16p requeues 0
marked 1357 early 35 pdrop 0 other 0
Signed-off-by: Eric Dumazet <eric.dumazet@...il.com>
---
include/linux/pkt_sched.h | 6 +-
include/net/red.h | 101 +++++++++++++++++++++++++++++-------
lib/reciprocal_div.c | 2
net/sched/sch_red.c | 21 +++++++
4 files changed, 111 insertions(+), 19 deletions(-)
diff --git a/include/linux/pkt_sched.h b/include/linux/pkt_sched.h
index fb556dc..e41e0d4 100644
--- a/include/linux/pkt_sched.h
+++ b/include/linux/pkt_sched.h
@@ -181,6 +181,7 @@ enum {
TCA_RED_UNSPEC,
TCA_RED_PARMS,
TCA_RED_STAB,
+ TCA_RED_MAX_P,
__TCA_RED_MAX,
};
@@ -194,8 +195,9 @@ struct tc_red_qopt {
unsigned char Plog; /* log(P_max/(qth_max-qth_min)) */
unsigned char Scell_log; /* cell size for idle damping */
unsigned char flags;
-#define TC_RED_ECN 1
-#define TC_RED_HARDDROP 2
+#define TC_RED_ECN 1
+#define TC_RED_HARDDROP 2
+#define TC_RED_ADAPTATIVE 4
};
struct tc_red_xstats {
diff --git a/include/net/red.h b/include/net/red.h
index b72a3b8..f4e9533 100644
--- a/include/net/red.h
+++ b/include/net/red.h
@@ -5,6 +5,7 @@
#include <net/pkt_sched.h>
#include <net/inet_ecn.h>
#include <net/dsfield.h>
+#include <linux/reciprocal_div.h>
/* Random Early Detection (RED) algorithm.
=======================================
@@ -87,6 +88,29 @@
etc.
*/
+/*
+ * Adaptative RED : An Algorithm for Increasing the Robustness of RED's AQM
+ * (Sally FLoyd, Ramakrishna Gummadi, and Scott Shenker) August 2001
+ *
+ * Every 500 ms:
+ * if (avg > target and max_p <= 0.5)
+ * increase max_p : max_p += alpha;
+ * else if (avg < target and max_p >= 0.01)
+ * decrease max_p : max_p *= beta;
+ *
+ * target :[qth_min + 0.4*(qth_min - qth_max),
+ * qth_min + 0.6*(qth_min - qth_max)].
+ * alpha : min(0.01, max_p / 4)
+ * beta : 0.9
+ * max_P is a Q0.32 fixed point number (with 32 bits mantissa)
+ * max_P between 0.01 and 0.5 (1% - 50%) [ Its no longer a negative power of two ]
+ */
+#define RED_ONE_PERCENT ((u32)DIV_ROUND_CLOSEST(1ULL<<32, 100))
+
+#define MAX_P_MIN (1 * RED_ONE_PERCENT)
+#define MAX_P_MAX (50 * RED_ONE_PERCENT)
+#define MAX_P_ALPHA(val) min(MAX_P_MIN, val / 4)
+
#define RED_STAB_SIZE 256
#define RED_STAB_MASK (RED_STAB_SIZE - 1)
@@ -101,10 +125,14 @@ struct red_stats {
struct red_parms {
/* Parameters */
- u32 qth_min; /* Min avg length threshold: A scaled */
- u32 qth_max; /* Max avg length threshold: A scaled */
+ u32 qth_min; /* Min avg length threshold: Wlog scaled */
+ u32 qth_max; /* Max avg length threshold: Wlog scaled */
u32 Scell_max;
- u32 Rmask; /* Cached random mask, see red_rmask */
+ u32 max_P; /* probability, [0 .. 1.0] 32 scaled */
+ u32 max_P_reciprocal; /* reciprocal_value(max_P / qth_delta) */
+ u32 qth_delta; /* max_th - min_th */
+ u32 target_min; /* min_th + 0.4*(max_th - min_th) */
+ u32 target_max; /* min_th + 0.6*(max_th - min_th) */
u8 Scell_log;
u8 Wlog; /* log(W) */
u8 Plog; /* random number bits */
@@ -115,19 +143,22 @@ struct red_parms {
number generation */
u32 qR; /* Cached random number */
- unsigned long qavg; /* Average queue length: A scaled */
+ unsigned long qavg; /* Average queue length: Wlog scaled */
ktime_t qidlestart; /* Start of current idle period */
};
-static inline u32 red_rmask(u8 Plog)
+static inline u32 red_maxp(u8 Plog)
{
- return Plog < 32 ? ((1 << Plog) - 1) : ~0UL;
+ return Plog < 32 ? (~0U >> Plog) : ~0U;
}
+
static inline void red_set_parms(struct red_parms *p,
u32 qth_min, u32 qth_max, u8 Wlog, u8 Plog,
u8 Scell_log, u8 *stab)
{
+ int delta = qth_max - qth_min;
+
/* Reset average queue length, the value is strictly bound
* to the parameters below, reseting hurts a bit but leaving
* it might result in an unreasonable qavg for a while. --TGR
@@ -139,14 +170,29 @@ static inline void red_set_parms(struct red_parms *p,
p->qth_max = qth_max << Wlog;
p->Wlog = Wlog;
p->Plog = Plog;
- p->Rmask = red_rmask(Plog);
+ if (delta < 0)
+ delta = 1;
+ p->qth_delta = delta;
+ p->max_P = red_maxp(Plog);
+ p->max_P *= delta; /* max_P = (qth_max-qth_min)/2^Plog */
+
+ p->max_P_reciprocal = reciprocal_value(p->max_P / delta);
+
+ /* RED Adaptative target :
+ * [min_th + 0.4*(min_th - max_th),
+ * min_th + 0.6*(min_th - max_th)].
+ */
+ delta /= 5;
+ p->target_min = qth_min + 2*delta;
+ p->target_max = qth_min + 3*delta;
+
p->Scell_log = Scell_log;
p->Scell_max = (255 << Scell_log);
memcpy(p->Stab, stab, sizeof(p->Stab));
}
-static inline int red_is_idling(struct red_parms *p)
+static inline int red_is_idling(const struct red_parms *p)
{
return p->qidlestart.tv64 != 0;
}
@@ -168,7 +214,7 @@ static inline void red_restart(struct red_parms *p)
p->qcount = -1;
}
-static inline unsigned long red_calc_qavg_from_idle_time(struct red_parms *p)
+static inline unsigned long red_calc_qavg_from_idle_time(const struct red_parms *p)
{
s64 delta = ktime_us_delta(ktime_get(), p->qidlestart);
long us_idle = min_t(s64, delta, p->Scell_max);
@@ -215,7 +261,7 @@ static inline unsigned long red_calc_qavg_from_idle_time(struct red_parms *p)
}
}
-static inline unsigned long red_calc_qavg_no_idle_time(struct red_parms *p,
+static inline unsigned long red_calc_qavg_no_idle_time(const struct red_parms *p,
unsigned int backlog)
{
/*
@@ -230,7 +276,7 @@ static inline unsigned long red_calc_qavg_no_idle_time(struct red_parms *p,
return p->qavg + (backlog - (p->qavg >> p->Wlog));
}
-static inline unsigned long red_calc_qavg(struct red_parms *p,
+static inline unsigned long red_calc_qavg(const struct red_parms *p,
unsigned int backlog)
{
if (!red_is_idling(p))
@@ -239,23 +285,24 @@ static inline unsigned long red_calc_qavg(struct red_parms *p,
return red_calc_qavg_from_idle_time(p);
}
-static inline u32 red_random(struct red_parms *p)
+
+static inline u32 red_random(const struct red_parms *p)
{
- return net_random() & p->Rmask;
+ return reciprocal_divide(net_random(), p->max_P_reciprocal);
}
-static inline int red_mark_probability(struct red_parms *p, unsigned long qavg)
+static inline int red_mark_probability(const struct red_parms *p, unsigned long qavg)
{
/* The formula used below causes questions.
- OK. qR is random number in the interval 0..Rmask
+ OK. qR is random number in the interval
+ (0..1/max_P)*(qth_max-qth_min)
i.e. 0..(2^Plog). If we used floating point
arithmetics, it would be: (2^Plog)*rnd_num,
where rnd_num is less 1.
Taking into account, that qavg have fixed
- point at Wlog, and Plog is related to max_P by
- max_P = (qth_max-qth_min)/2^Plog; two lines
+ point at Wlog, two lines
below have the following floating point equivalent:
max_P*(qavg - qth_min)/(qth_max-qth_min) < rnd/qcount
@@ -315,4 +362,24 @@ static inline int red_action(struct red_parms *p, unsigned long qavg)
return RED_DONT_MARK;
}
+static inline void red_adaptative_algo(struct red_parms *p)
+{
+ unsigned long qavg;
+ u32 max_p_delta;
+
+ qavg = p->qavg;
+ if (red_is_idling(p))
+ qavg = red_calc_qavg_from_idle_time(p);
+
+ /* p->qavg is fixed point number with point at Wlog */
+ qavg >>= p->Wlog;
+
+ if (qavg > p->target_max && p->max_P <= MAX_P_MAX)
+ p->max_P += MAX_P_ALPHA(p->max_P); /* maxp = maxp + alpha */
+ else if (qavg < p->target_min && p->max_P >= MAX_P_MIN)
+ p->max_P = (p->max_P/10)*9; /* maxp = maxp * Beta */
+
+ max_p_delta = DIV_ROUND_CLOSEST(p->max_P, p->qth_delta);
+ p->max_P_reciprocal = reciprocal_value(max_p_delta);
+}
#endif
diff --git a/lib/reciprocal_div.c b/lib/reciprocal_div.c
index 6a3bd48..75510e9 100644
--- a/lib/reciprocal_div.c
+++ b/lib/reciprocal_div.c
@@ -1,5 +1,6 @@
#include <asm/div64.h>
#include <linux/reciprocal_div.h>
+#include <linux/export.h>
u32 reciprocal_value(u32 k)
{
@@ -7,3 +8,4 @@ u32 reciprocal_value(u32 k)
do_div(val, k);
return (u32)val;
}
+EXPORT_SYMBOL(reciprocal_value);
diff --git a/net/sched/sch_red.c b/net/sched/sch_red.c
index d617161..8f5a85b 100644
--- a/net/sched/sch_red.c
+++ b/net/sched/sch_red.c
@@ -39,6 +39,7 @@
struct red_sched_data {
u32 limit; /* HARD maximal queue length */
unsigned char flags;
+ struct timer_list adapt_timer;
struct red_parms parms;
struct red_stats stats;
struct Qdisc *qdisc;
@@ -161,6 +162,8 @@ static void red_reset(struct Qdisc *sch)
static void red_destroy(struct Qdisc *sch)
{
struct red_sched_data *q = qdisc_priv(sch);
+
+ del_timer_sync(&q->adapt_timer);
qdisc_destroy(q->qdisc);
}
@@ -209,6 +212,10 @@ static int red_change(struct Qdisc *sch, struct nlattr *opt)
ctl->Plog, ctl->Scell_log,
nla_data(tb[TCA_RED_STAB]));
+ del_timer(&q->adapt_timer);
+ if (ctl->flags & TC_RED_ADAPTATIVE)
+ mod_timer(&q->adapt_timer, jiffies + HZ/2);
+
if (!q->qdisc->q.qlen)
red_start_of_idle_period(&q->parms);
@@ -216,11 +223,24 @@ static int red_change(struct Qdisc *sch, struct nlattr *opt)
return 0;
}
+static inline void red_adaptative_timer(unsigned long arg)
+{
+ struct Qdisc *sch = (struct Qdisc *)arg;
+ struct red_sched_data *q = qdisc_priv(sch);
+ spinlock_t *root_lock = qdisc_lock(qdisc_root_sleeping(sch));
+
+ spin_lock(root_lock);
+ red_adaptative_algo(&q->parms);
+ mod_timer(&q->adapt_timer, jiffies + HZ/2);
+ spin_unlock(root_lock);
+}
+
static int red_init(struct Qdisc *sch, struct nlattr *opt)
{
struct red_sched_data *q = qdisc_priv(sch);
q->qdisc = &noop_qdisc;
+ setup_timer(&q->adapt_timer, red_adaptative_timer, (unsigned long)sch);
return red_change(sch, opt);
}
@@ -243,6 +263,7 @@ static int red_dump(struct Qdisc *sch, struct sk_buff *skb)
if (opts == NULL)
goto nla_put_failure;
NLA_PUT(skb, TCA_RED_PARMS, sizeof(opt), &opt);
+ NLA_PUT_U32(skb, TCA_RED_MAX_P, q->parms.max_P);
return nla_nest_end(skb, opts);
nla_put_failure:
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists