Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * ip_vs_est.c: simple rate estimator for IPVS |
| 3 | * |
| 4 | * Version: $Id: ip_vs_est.c,v 1.4 2002/11/30 01:50:35 wensong Exp $ |
| 5 | * |
| 6 | * Authors: Wensong Zhang <wensong@linuxvirtualserver.org> |
| 7 | * |
| 8 | * This program is free software; you can redistribute it and/or |
| 9 | * modify it under the terms of the GNU General Public License |
| 10 | * as published by the Free Software Foundation; either version |
| 11 | * 2 of the License, or (at your option) any later version. |
| 12 | * |
| 13 | * Changes: |
| 14 | * |
| 15 | */ |
| 16 | #include <linux/kernel.h> |
| 17 | #include <linux/types.h> |
| 18 | |
| 19 | #include <net/ip_vs.h> |
| 20 | |
| 21 | /* |
| 22 | This code is to estimate rate in a shorter interval (such as 8 |
| 23 | seconds) for virtual services and real servers. For measure rate in a |
| 24 | long interval, it is easy to implement a user level daemon which |
| 25 | periodically reads those statistical counters and measure rate. |
| 26 | |
| 27 | Currently, the measurement is activated by slow timer handler. Hope |
| 28 | this measurement will not introduce too much load. |
| 29 | |
| 30 | We measure rate during the last 8 seconds every 2 seconds: |
| 31 | |
| 32 | avgrate = avgrate*(1-W) + rate*W |
| 33 | |
| 34 | where W = 2^(-2) |
| 35 | |
| 36 | NOTES. |
| 37 | |
| 38 | * The stored value for average bps is scaled by 2^5, so that maximal |
| 39 | rate is ~2.15Gbits/s, average pps and cps are scaled by 2^10. |
| 40 | |
| 41 | * A lot code is taken from net/sched/estimator.c |
| 42 | */ |
| 43 | |
| 44 | |
| 45 | struct ip_vs_estimator |
| 46 | { |
| 47 | struct ip_vs_estimator *next; |
| 48 | struct ip_vs_stats *stats; |
| 49 | |
| 50 | u32 last_conns; |
| 51 | u32 last_inpkts; |
| 52 | u32 last_outpkts; |
| 53 | u64 last_inbytes; |
| 54 | u64 last_outbytes; |
| 55 | |
| 56 | u32 cps; |
| 57 | u32 inpps; |
| 58 | u32 outpps; |
| 59 | u32 inbps; |
| 60 | u32 outbps; |
| 61 | }; |
| 62 | |
| 63 | |
| 64 | static struct ip_vs_estimator *est_list = NULL; |
| 65 | static DEFINE_RWLOCK(est_lock); |
| 66 | static struct timer_list est_timer; |
| 67 | |
| 68 | static void estimation_timer(unsigned long arg) |
| 69 | { |
| 70 | struct ip_vs_estimator *e; |
| 71 | struct ip_vs_stats *s; |
| 72 | u32 n_conns; |
| 73 | u32 n_inpkts, n_outpkts; |
| 74 | u64 n_inbytes, n_outbytes; |
| 75 | u32 rate; |
| 76 | |
| 77 | read_lock(&est_lock); |
| 78 | for (e = est_list; e; e = e->next) { |
| 79 | s = e->stats; |
| 80 | |
| 81 | spin_lock(&s->lock); |
| 82 | n_conns = s->conns; |
| 83 | n_inpkts = s->inpkts; |
| 84 | n_outpkts = s->outpkts; |
| 85 | n_inbytes = s->inbytes; |
| 86 | n_outbytes = s->outbytes; |
| 87 | |
| 88 | /* scaled by 2^10, but divided 2 seconds */ |
| 89 | rate = (n_conns - e->last_conns)<<9; |
| 90 | e->last_conns = n_conns; |
| 91 | e->cps += ((long)rate - (long)e->cps)>>2; |
| 92 | s->cps = (e->cps+0x1FF)>>10; |
| 93 | |
| 94 | rate = (n_inpkts - e->last_inpkts)<<9; |
| 95 | e->last_inpkts = n_inpkts; |
| 96 | e->inpps += ((long)rate - (long)e->inpps)>>2; |
| 97 | s->inpps = (e->inpps+0x1FF)>>10; |
| 98 | |
| 99 | rate = (n_outpkts - e->last_outpkts)<<9; |
| 100 | e->last_outpkts = n_outpkts; |
| 101 | e->outpps += ((long)rate - (long)e->outpps)>>2; |
| 102 | s->outpps = (e->outpps+0x1FF)>>10; |
| 103 | |
| 104 | rate = (n_inbytes - e->last_inbytes)<<4; |
| 105 | e->last_inbytes = n_inbytes; |
| 106 | e->inbps += ((long)rate - (long)e->inbps)>>2; |
| 107 | s->inbps = (e->inbps+0xF)>>5; |
| 108 | |
| 109 | rate = (n_outbytes - e->last_outbytes)<<4; |
| 110 | e->last_outbytes = n_outbytes; |
| 111 | e->outbps += ((long)rate - (long)e->outbps)>>2; |
| 112 | s->outbps = (e->outbps+0xF)>>5; |
| 113 | spin_unlock(&s->lock); |
| 114 | } |
| 115 | read_unlock(&est_lock); |
| 116 | mod_timer(&est_timer, jiffies + 2*HZ); |
| 117 | } |
| 118 | |
| 119 | int ip_vs_new_estimator(struct ip_vs_stats *stats) |
| 120 | { |
| 121 | struct ip_vs_estimator *est; |
| 122 | |
| 123 | est = kmalloc(sizeof(*est), GFP_KERNEL); |
| 124 | if (est == NULL) |
| 125 | return -ENOMEM; |
| 126 | |
| 127 | memset(est, 0, sizeof(*est)); |
| 128 | est->stats = stats; |
| 129 | est->last_conns = stats->conns; |
| 130 | est->cps = stats->cps<<10; |
| 131 | |
| 132 | est->last_inpkts = stats->inpkts; |
| 133 | est->inpps = stats->inpps<<10; |
| 134 | |
| 135 | est->last_outpkts = stats->outpkts; |
| 136 | est->outpps = stats->outpps<<10; |
| 137 | |
| 138 | est->last_inbytes = stats->inbytes; |
| 139 | est->inbps = stats->inbps<<5; |
| 140 | |
| 141 | est->last_outbytes = stats->outbytes; |
| 142 | est->outbps = stats->outbps<<5; |
| 143 | |
| 144 | write_lock_bh(&est_lock); |
| 145 | est->next = est_list; |
| 146 | if (est->next == NULL) { |
| 147 | init_timer(&est_timer); |
| 148 | est_timer.expires = jiffies + 2*HZ; |
| 149 | est_timer.function = estimation_timer; |
| 150 | add_timer(&est_timer); |
| 151 | } |
| 152 | est_list = est; |
| 153 | write_unlock_bh(&est_lock); |
| 154 | return 0; |
| 155 | } |
| 156 | |
| 157 | void ip_vs_kill_estimator(struct ip_vs_stats *stats) |
| 158 | { |
| 159 | struct ip_vs_estimator *est, **pest; |
| 160 | int killed = 0; |
| 161 | |
| 162 | write_lock_bh(&est_lock); |
| 163 | pest = &est_list; |
| 164 | while ((est=*pest) != NULL) { |
| 165 | if (est->stats != stats) { |
| 166 | pest = &est->next; |
| 167 | continue; |
| 168 | } |
| 169 | *pest = est->next; |
| 170 | kfree(est); |
| 171 | killed++; |
| 172 | } |
| 173 | if (killed && est_list == NULL) |
| 174 | del_timer_sync(&est_timer); |
| 175 | write_unlock_bh(&est_lock); |
| 176 | } |
| 177 | |
| 178 | void ip_vs_zero_estimator(struct ip_vs_stats *stats) |
| 179 | { |
| 180 | struct ip_vs_estimator *e; |
| 181 | |
| 182 | write_lock_bh(&est_lock); |
| 183 | for (e = est_list; e; e = e->next) { |
| 184 | if (e->stats != stats) |
| 185 | continue; |
| 186 | |
| 187 | /* set counters zero */ |
| 188 | e->last_conns = 0; |
| 189 | e->last_inpkts = 0; |
| 190 | e->last_outpkts = 0; |
| 191 | e->last_inbytes = 0; |
| 192 | e->last_outbytes = 0; |
| 193 | e->cps = 0; |
| 194 | e->inpps = 0; |
| 195 | e->outpps = 0; |
| 196 | e->inbps = 0; |
| 197 | e->outbps = 0; |
| 198 | } |
| 199 | write_unlock_bh(&est_lock); |
| 200 | } |