Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * ip_vs_est.c: simple rate estimator for IPVS | |
3 | * | |
1da177e4 LT |
4 | * Authors: Wensong Zhang <wensong@linuxvirtualserver.org> |
5 | * | |
6 | * This program is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public License | |
8 | * as published by the Free Software Foundation; either version | |
9 | * 2 of the License, or (at your option) any later version. | |
10 | * | |
11 | * Changes: | |
12 | * | |
13 | */ | |
14 | #include <linux/kernel.h> | |
14c85021 ACM |
15 | #include <linux/jiffies.h> |
16 | #include <linux/slab.h> | |
1da177e4 | 17 | #include <linux/types.h> |
4ffd2e49 | 18 | #include <linux/interrupt.h> |
90754f8e | 19 | #include <linux/sysctl.h> |
3a14a313 | 20 | #include <linux/list.h> |
1da177e4 LT |
21 | |
22 | #include <net/ip_vs.h> | |
23 | ||
24 | /* | |
25 | This code is to estimate rate in a shorter interval (such as 8 | |
26 | seconds) for virtual services and real servers. For measure rate in a | |
27 | long interval, it is easy to implement a user level daemon which | |
28 | periodically reads those statistical counters and measure rate. | |
29 | ||
30 | Currently, the measurement is activated by slow timer handler. Hope | |
31 | this measurement will not introduce too much load. | |
32 | ||
33 | We measure rate during the last 8 seconds every 2 seconds: | |
34 | ||
35 | avgrate = avgrate*(1-W) + rate*W | |
36 | ||
37 | where W = 2^(-2) | |
38 | ||
39 | NOTES. | |
40 | ||
41 | * The stored value for average bps is scaled by 2^5, so that maximal | |
42 | rate is ~2.15Gbits/s, average pps and cps are scaled by 2^10. | |
43 | ||
44 | * A lot code is taken from net/sched/estimator.c | |
45 | */ | |
46 | ||
47 | ||
3a14a313 | 48 | static void estimation_timer(unsigned long arg); |
1da177e4 | 49 | |
3a14a313 SW |
50 | static LIST_HEAD(est_list); |
51 | static DEFINE_SPINLOCK(est_lock); | |
52 | static DEFINE_TIMER(est_timer, estimation_timer, 0, 0); | |
1da177e4 LT |
53 | |
54 | static void estimation_timer(unsigned long arg) | |
55 | { | |
56 | struct ip_vs_estimator *e; | |
57 | struct ip_vs_stats *s; | |
58 | u32 n_conns; | |
59 | u32 n_inpkts, n_outpkts; | |
60 | u64 n_inbytes, n_outbytes; | |
61 | u32 rate; | |
62 | ||
3a14a313 SW |
63 | spin_lock(&est_lock); |
64 | list_for_each_entry(e, &est_list, list) { | |
65 | s = container_of(e, struct ip_vs_stats, est); | |
1da177e4 LT |
66 | |
67 | spin_lock(&s->lock); | |
68 | n_conns = s->conns; | |
69 | n_inpkts = s->inpkts; | |
70 | n_outpkts = s->outpkts; | |
71 | n_inbytes = s->inbytes; | |
72 | n_outbytes = s->outbytes; | |
73 | ||
74 | /* scaled by 2^10, but divided 2 seconds */ | |
75 | rate = (n_conns - e->last_conns)<<9; | |
76 | e->last_conns = n_conns; | |
77 | e->cps += ((long)rate - (long)e->cps)>>2; | |
78 | s->cps = (e->cps+0x1FF)>>10; | |
79 | ||
80 | rate = (n_inpkts - e->last_inpkts)<<9; | |
81 | e->last_inpkts = n_inpkts; | |
82 | e->inpps += ((long)rate - (long)e->inpps)>>2; | |
83 | s->inpps = (e->inpps+0x1FF)>>10; | |
84 | ||
85 | rate = (n_outpkts - e->last_outpkts)<<9; | |
86 | e->last_outpkts = n_outpkts; | |
87 | e->outpps += ((long)rate - (long)e->outpps)>>2; | |
88 | s->outpps = (e->outpps+0x1FF)>>10; | |
89 | ||
90 | rate = (n_inbytes - e->last_inbytes)<<4; | |
91 | e->last_inbytes = n_inbytes; | |
92 | e->inbps += ((long)rate - (long)e->inbps)>>2; | |
93 | s->inbps = (e->inbps+0xF)>>5; | |
94 | ||
95 | rate = (n_outbytes - e->last_outbytes)<<4; | |
96 | e->last_outbytes = n_outbytes; | |
97 | e->outbps += ((long)rate - (long)e->outbps)>>2; | |
98 | s->outbps = (e->outbps+0xF)>>5; | |
99 | spin_unlock(&s->lock); | |
100 | } | |
3a14a313 | 101 | spin_unlock(&est_lock); |
1da177e4 LT |
102 | mod_timer(&est_timer, jiffies + 2*HZ); |
103 | } | |
104 | ||
3a14a313 | 105 | void ip_vs_new_estimator(struct ip_vs_stats *stats) |
1da177e4 | 106 | { |
3a14a313 | 107 | struct ip_vs_estimator *est = &stats->est; |
1da177e4 | 108 | |
3a14a313 | 109 | INIT_LIST_HEAD(&est->list); |
1da177e4 | 110 | |
1da177e4 LT |
111 | est->last_conns = stats->conns; |
112 | est->cps = stats->cps<<10; | |
113 | ||
114 | est->last_inpkts = stats->inpkts; | |
115 | est->inpps = stats->inpps<<10; | |
116 | ||
117 | est->last_outpkts = stats->outpkts; | |
118 | est->outpps = stats->outpps<<10; | |
119 | ||
120 | est->last_inbytes = stats->inbytes; | |
121 | est->inbps = stats->inbps<<5; | |
122 | ||
123 | est->last_outbytes = stats->outbytes; | |
124 | est->outbps = stats->outbps<<5; | |
125 | ||
3a14a313 SW |
126 | spin_lock_bh(&est_lock); |
127 | if (list_empty(&est_list)) | |
128 | mod_timer(&est_timer, jiffies + 2 * HZ); | |
129 | list_add(&est->list, &est_list); | |
130 | spin_unlock_bh(&est_lock); | |
1da177e4 LT |
131 | } |
132 | ||
133 | void ip_vs_kill_estimator(struct ip_vs_stats *stats) | |
134 | { | |
3a14a313 SW |
135 | struct ip_vs_estimator *est = &stats->est; |
136 | ||
137 | spin_lock_bh(&est_lock); | |
138 | list_del(&est->list); | |
139 | while (list_empty(&est_list) && try_to_del_timer_sync(&est_timer) < 0) { | |
140 | spin_unlock_bh(&est_lock); | |
8ab19ea3 | 141 | cpu_relax(); |
3a14a313 | 142 | spin_lock_bh(&est_lock); |
8ab19ea3 | 143 | } |
3a14a313 | 144 | spin_unlock_bh(&est_lock); |
1da177e4 LT |
145 | } |
146 | ||
147 | void ip_vs_zero_estimator(struct ip_vs_stats *stats) | |
148 | { | |
3a14a313 SW |
149 | struct ip_vs_estimator *est = &stats->est; |
150 | ||
151 | /* set counters zero, caller must hold the stats->lock lock */ | |
152 | est->last_inbytes = 0; | |
153 | est->last_outbytes = 0; | |
154 | est->last_conns = 0; | |
155 | est->last_inpkts = 0; | |
156 | est->last_outpkts = 0; | |
157 | est->cps = 0; | |
158 | est->inpps = 0; | |
159 | est->outpps = 0; | |
160 | est->inbps = 0; | |
161 | est->outbps = 0; | |
1da177e4 | 162 | } |