blob: 5d3ae03e22a70fdf49350f833a282c0a3e12624a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * net/sched/estimator.c Simple rate estimator.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
8 *
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10 */
11
12#include <asm/uaccess.h>
13#include <asm/system.h>
14#include <linux/bitops.h>
15#include <linux/module.h>
16#include <linux/types.h>
17#include <linux/kernel.h>
18#include <linux/jiffies.h>
19#include <linux/string.h>
20#include <linux/mm.h>
21#include <linux/socket.h>
22#include <linux/sockios.h>
23#include <linux/in.h>
24#include <linux/errno.h>
25#include <linux/interrupt.h>
26#include <linux/netdevice.h>
27#include <linux/skbuff.h>
28#include <linux/rtnetlink.h>
29#include <linux/init.h>
30#include <net/sock.h>
31#include <net/pkt_sched.h>
32
33/*
34 This code is NOT intended to be used for statistics collection,
35 its purpose is to provide a base for statistical multiplexing
36 for controlled load service.
37 If you need only statistics, run a user level daemon which
38 periodically reads byte counters.
39
40 Unfortunately, rate estimation is not a very easy task.
41 F.e. I did not find a simple way to estimate the current peak rate
42 and even failed to formulate the problem 8)8)
43
44 So I preferred not to built an estimator into the scheduler,
45 but run this task separately.
46 Ideally, it should be kernel thread(s), but for now it runs
47 from timers, which puts apparent top bounds on the number of rated
48 flows, has minimal overhead on small, but is enough
49 to handle controlled load service, sets of aggregates.
50
51 We measure rate over A=(1<<interval) seconds and evaluate EWMA:
52
53 avrate = avrate*(1-W) + rate*W
54
55 where W is chosen as negative power of 2: W = 2^(-ewma_log)
56
57 The resulting time constant is:
58
59 T = A/(-ln(1-W))
60
61
62 NOTES.
63
64 * The stored value for avbps is scaled by 2^5, so that maximal
65 rate is ~1Gbit, avpps is scaled by 2^10.
66
67 * Minimal interval is HZ/4=250msec (it is the greatest common divisor
68 for HZ=100 and HZ=1024 8)), maximal interval
69 is (HZ*2^EST_MAX_INTERVAL)/4 = 8sec. Shorter intervals
70 are too expensive, longer ones can be implemented
71 at user level painlessly.
72 */
73
74#define EST_MAX_INTERVAL 5
75
76struct qdisc_estimator
77{
78 struct qdisc_estimator *next;
79 struct tc_stats *stats;
80 spinlock_t *stats_lock;
81 unsigned interval;
82 int ewma_log;
83 u64 last_bytes;
84 u32 last_packets;
85 u32 avpps;
86 u32 avbps;
87};
88
89struct qdisc_estimator_head
90{
91 struct timer_list timer;
92 struct qdisc_estimator *list;
93};
94
95static struct qdisc_estimator_head elist[EST_MAX_INTERVAL+1];
96
97/* Estimator array lock */
98static DEFINE_RWLOCK(est_lock);
99
100static void est_timer(unsigned long arg)
101{
102 int idx = (int)arg;
103 struct qdisc_estimator *e;
104
105 read_lock(&est_lock);
106 for (e = elist[idx].list; e; e = e->next) {
107 struct tc_stats *st = e->stats;
108 u64 nbytes;
109 u32 npackets;
110 u32 rate;
111
112 spin_lock(e->stats_lock);
113 nbytes = st->bytes;
114 npackets = st->packets;
115 rate = (nbytes - e->last_bytes)<<(7 - idx);
116 e->last_bytes = nbytes;
117 e->avbps += ((long)rate - (long)e->avbps) >> e->ewma_log;
118 st->bps = (e->avbps+0xF)>>5;
119
120 rate = (npackets - e->last_packets)<<(12 - idx);
121 e->last_packets = npackets;
122 e->avpps += ((long)rate - (long)e->avpps) >> e->ewma_log;
123 e->stats->pps = (e->avpps+0x1FF)>>10;
124 spin_unlock(e->stats_lock);
125 }
126
127 mod_timer(&elist[idx].timer, jiffies + ((HZ<<idx)/4));
128 read_unlock(&est_lock);
129}
130
131int qdisc_new_estimator(struct tc_stats *stats, spinlock_t *stats_lock, struct rtattr *opt)
132{
133 struct qdisc_estimator *est;
134 struct tc_estimator *parm = RTA_DATA(opt);
135
136 if (RTA_PAYLOAD(opt) < sizeof(*parm))
137 return -EINVAL;
138
139 if (parm->interval < -2 || parm->interval > 3)
140 return -EINVAL;
141
142 est = kmalloc(sizeof(*est), GFP_KERNEL);
143 if (est == NULL)
144 return -ENOBUFS;
145
146 memset(est, 0, sizeof(*est));
147 est->interval = parm->interval + 2;
148 est->stats = stats;
149 est->stats_lock = stats_lock;
150 est->ewma_log = parm->ewma_log;
151 est->last_bytes = stats->bytes;
152 est->avbps = stats->bps<<5;
153 est->last_packets = stats->packets;
154 est->avpps = stats->pps<<10;
155
156 est->next = elist[est->interval].list;
157 if (est->next == NULL) {
158 init_timer(&elist[est->interval].timer);
159 elist[est->interval].timer.data = est->interval;
160 elist[est->interval].timer.expires = jiffies + ((HZ<<est->interval)/4);
161 elist[est->interval].timer.function = est_timer;
162 add_timer(&elist[est->interval].timer);
163 }
164 write_lock_bh(&est_lock);
165 elist[est->interval].list = est;
166 write_unlock_bh(&est_lock);
167 return 0;
168}
169
170void qdisc_kill_estimator(struct tc_stats *stats)
171{
172 int idx;
173 struct qdisc_estimator *est, **pest;
174
175 for (idx=0; idx <= EST_MAX_INTERVAL; idx++) {
176 int killed = 0;
177 pest = &elist[idx].list;
178 while ((est=*pest) != NULL) {
179 if (est->stats != stats) {
180 pest = &est->next;
181 continue;
182 }
183
184 write_lock_bh(&est_lock);
185 *pest = est->next;
186 write_unlock_bh(&est_lock);
187
188 kfree(est);
189 killed++;
190 }
191 if (killed && elist[idx].list == NULL)
192 del_timer(&elist[idx].timer);
193 }
194}
195
196EXPORT_SYMBOL(qdisc_kill_estimator);
197EXPORT_SYMBOL(qdisc_new_estimator);