1 /* See the DRL-LICENSE file for this file's software license. */
4 * Thread to periodically calculate the estimated local limits
5 * Barath Raghavan 2006/2007
12 /** The size of the buffer we use to hold tc commands. */
13 #define CMD_BUFFER_SIZE 200
16 #include "raterouter.h"
18 #include "ratetypes.h" /* needs util and pthread.h */
21 static int underlimit_flowcount_count = 0;
22 static int underlimit_normal_count = 0;
23 extern uint8_t system_loglevel;
26 * Called for each identity each estimate interval. Uses flow table information
27 * to estimate the current aggregate rate and the rate of the individual flows
30 static void estimate(identity_t *ident) {
33 gettimeofday(&now, NULL);
35 pthread_mutex_lock(&ident->table_mutex); /* CLUNK ! */
37 ident->table_update_function(ident->table, now, ident->ewma_weight);
39 pthread_mutex_unlock(&ident->table_mutex); /* CLINK ! */
43 * Determines the FPS weight allocation when the identity is under its current
46 static double allocate_fps_under_limit(identity_t *ident, uint32_t local_rate, double peer_weights) {
47 uint32_t target = local_rate;
49 double total_weight = peer_weights + ident->last_localweight;
51 if (ident->flowstart) {
52 target = local_rate*4;
53 if (local_rate >= FLOW_START_THRESHOLD) {
54 ident->flowstart = false;
58 /* June 16, 2008 (KCW)
59 * ident->flowstart gets set initially to one, but it is never set again. However,
60 * if a limiter gets flows and then the number of flows drops to zero, it has trouble
61 * increasing the limit again. */
62 if (local_rate < FLOW_START_THRESHOLD) {
63 ident->flowstart = true;
67 if (target >= ident->limit) {
68 ideal_weight = total_weight;
69 } else if (target <= 0) {
70 ideal_weight = 0; // no flows here
72 ideal_weight = ((double)target / (double)ident->limit) * total_weight;
76 else if (peer_weights <= 0) {
78 // doesn't matter what we pick as our weight, so pick 1 / N.
79 ideal_weight = MAX_FLOW_SCALING_FACTOR / (remote_count(ident->i_handle) + 1);
81 ideal_weight = ((double)target / (double)ident->limit) * total_weight;
84 double divisor = (double) ident->limit - (double) target;
85 ideal_weight = ((double) target * peer_weights) / divisor;
87 ideal_weight = ((double)target / (double)ident->limit) * total_weight;
96 * Determines the FPS weight allocation when the identity is over its current
99 static double allocate_fps_over_limit(identity_t *ident) {
102 if (ident->common.max_flow_rate > 0) {
103 ideal_weight = (double) ident->locallimit / (double) ident->common.max_flow_rate;
105 printlog(LOG_DEBUG, "%.3f %d %d FlowCount, TotalRate, MaxRate\n",
106 ideal_weight, ident->common.rate, ident->common.max_flow_rate);
115 * Determines the amount of FPS weight to allocate to the identity during each
116 * estimate interval. Note that total_weight includes local weight.
118 static uint32_t allocate_fps(identity_t *ident, double total_weight) {
119 common_accounting_t *ftable = &ident->common; /* Common flow table info */
120 uint32_t local_rate = ftable->rate;
121 uint32_t ideallocal = 0;
122 double peer_weights; /* sum of weights of all other limiters */
123 double idealweight = 0;
124 double last_portion = 0;
125 double this_portion = 0;
127 static int dampen = 0;
128 int dampen_increase = 0;
130 double ideal_under = 0;
131 double ideal_over = 0;
136 1. the aggregate is < limit
137 2. the aggregate is >= limit
139 peer_weights = total_weight - ident->last_localweight;
140 if (peer_weights < 0) {
146 (int64_t) ftable->inst_rate - (int64_t) ftable->last_inst_rate;
148 (double) ident->limit * (double) LARGE_INCREASE_PERCENTAGE / 10;
150 if (rate_delta > threshold) {
152 printlog(LOG_DEBUG, "DAMPEN: delta(%.3f) thresh(%.3f)\n",
153 rate_delta, threshold);
157 if (local_rate <= 0) {
159 } else if (dampen_increase == 0 && (ident->locallimit <= 0 || local_rate < ident->locallimit || ident->flowstart)) {
160 /* We're under the limit - all flows are bottlenecked. */
161 idealweight = allocate_fps_under_limit(ident, local_rate, peer_weights);
162 ideal_over = allocate_fps_over_limit(ident);
163 ideal_under = idealweight;
165 if (ideal_over < idealweight) {
166 idealweight = ideal_over;
169 underlimit_flowcount_count += 1;
173 underlimit_normal_count += 1;
177 ident->localweight = (ident->localweight * ident->ewma_weight +
178 idealweight * (1 - ident->ewma_weight));
181 idealweight = allocate_fps_over_limit(ident);
184 ident->localweight = (ident->localweight * ident->ewma_weight +
185 idealweight * (1 - ident->ewma_weight));
187 /* This is the portion of the total weight in the system that was caused
188 * by this limiter in the last interval. */
189 last_portion = ident->last_localweight / total_weight;
191 /* This is the fraction of the total weight in the system that our
192 * proposed value for idealweight would use. */
193 this_portion = ident->localweight / (peer_weights + ident->localweight);
195 /* Dampen the large increase the first time... */
196 if (dampen == 0 && (this_portion - last_portion > LARGE_INCREASE_PERCENTAGE)) {
197 ident->localweight = ident->last_localweight + (LARGE_INCREASE_PERCENTAGE * total_weight);
203 ideal_under = allocate_fps_under_limit(ident, local_rate, peer_weights);
204 ideal_over = idealweight;
209 /* Convert weight into a rate - add in our new local weight */
210 ident->total_weight = total_weight = ident->localweight + peer_weights;
212 /* compute local allocation:
213 if there is traffic elsewhere, use the weights
214 otherwise do a L/n allocation */
215 if (total_weight > 0) {
216 //if (peer_weights > 0) {
217 ideallocal = (uint32_t) (ident->localweight * ident->limit / total_weight);
219 ideallocal = ident->limit / (ident->comm.remote_node_count + 1);
222 printlog(LOG_DEBUG, "%.3f ActualWeight\n", ident->localweight);
224 printlog(LOG_DEBUG, "%.3f %.3f %.3f %.3f Under / Over / Actual / Rate\n",
225 ideal_under / (ideal_under + peer_weights),
226 ideal_over / (ideal_over + peer_weights),
227 ident->localweight / (ident->localweight + peer_weights),
228 (double) local_rate / (double) ident->limit);
230 printlog(LOG_DEBUG, "%.3f %.3f IdealUnd IdealOve\n",ideal_under,ideal_over);
232 if (system_loglevel == LOG_DEBUG) {
233 printf("local_rate: %d, idealweight: %.3f, localweight: %.3f, total_weight: %.3f\n",
234 local_rate, idealweight, ident->localweight, total_weight);
237 //printf("Dampen: %d, dampen_increase: %d, peer_weights: %.3f, regime: %d\n",
238 // dampen, dampen_increase, peer_weights, regime);
240 //printf("normal_count: %d, flowcount_count: %d\n", underlimit_normal_count, underlimit_flowcount_count);
243 printlog(LOG_DEBUG, "MIN: min said to use flow counting, which was %.3f when other method said %.3f.\n",
244 ideal_over, ideal_under);
247 printlog(LOG_DEBUG, "ideallocal is %d\n", ideallocal);
253 * Determines the local drop probability for a GRD identity every estimate
256 static double allocate_grd(identity_t *ident, double aggdemand) {
258 double global_limit = (double) (ident->limit);
260 if (aggdemand > global_limit) {
261 dropprob = (aggdemand-global_limit)/aggdemand;
266 if (system_loglevel == LOG_DEBUG) {
267 printf("local rate: %d, aggregate demand: %.3f, drop prob: %.3f\n",
268 ident->common.rate, aggdemand, dropprob);
275 * Given current estimates of local rate (weight) and remote rates (weights)
276 * use GRD or FPS to calculate a new local limit.
278 static void allocate(limiter_t *limiter, identity_t *ident) {
279 /* Represents aggregate rate for GRD and aggregate weight for FPS. */
282 /* Read comm_val from comm layer. */
283 if (limiter->policy == POLICY_FPS) {
284 read_comm(&ident->comm, &comm_val,
285 ident->total_weight / (double) (ident->comm.remote_node_count + 1));
287 read_comm(&ident->comm, &comm_val,
288 (double) (ident->limit / (double) (ident->comm.remote_node_count + 1)));
290 printlog(LOG_DEBUG, "%.3f Aggregate weight/rate (FPS/GRD)\n", comm_val);
292 /* Experimental printing. */
293 printlog(LOG_DEBUG, "%.3f \t Kbps used rate. ID:%d\n",
294 (double) ident->common.rate / (double) 128, ident->id);
295 ident->avg_bytes += ident->common.rate;
297 if (limiter->policy == POLICY_FPS) {
298 ident->locallimit = allocate_fps(ident, comm_val);
299 ident->last_localweight = ident->localweight;
301 /* Update other limiters with our weight by writing to comm layer. */
302 write_local_value(&ident->comm, ident->localweight);
304 ident->locallimit = 0; /* Unused with GRD. */
305 ident->last_drop_prob = ident->drop_prob;
306 ident->drop_prob = allocate_grd(ident, comm_val);
308 /* Update other limiters with our rate by writing to comm layer. */
309 write_local_value(&ident->comm, ident->common.rate);
312 /* Update identity state. */
313 ident->common.last_rate = ident->common.rate;
317 * Traces all of the parent pointers of a leaf all the way to the root in
318 * order to find the maximum drop probability in the chain.
320 static double find_leaf_drop_prob(leaf_t *leaf) {
321 identity_t *current = leaf->parent;
326 while (current != NULL) {
327 if (current->drop_prob > result) {
328 result = current->drop_prob;
330 current = current->parent;
337 * This is called once per estimate interval to enforce the rate that allocate
338 * has decided upon. It makes calls to tc using system().
340 static void enforce(limiter_t *limiter, identity_t *ident) {
341 char cmd[CMD_BUFFER_SIZE];
345 switch (limiter->policy) {
348 /* TC treats limits of 0 (8bit) as unlimited, which causes the
349 * entire rate limiting system to become unpredictable. In
350 * reality, we also don't want any limiter to be able to set its
351 * limit so low that it chokes all of the flows to the point that
352 * they can't increase. Thus, when we're setting a low limit, we
353 * make sure that it isn't too low by using the
354 * FLOW_START_THRESHOLD. */
356 if (ident->locallimit < FLOW_START_THRESHOLD) {
357 ident->locallimit = FLOW_START_THRESHOLD * 2;
360 /* Do not allow the node to set a limit higher than its
361 * administratively assigned upper limit (bwcap). */
362 if (limiter->nodelimit != 0 && ident->locallimit > limiter->nodelimit) {
363 ident->locallimit = limiter->nodelimit;
366 if (system_loglevel == LOG_DEBUG) {
367 printf("FPS: Setting local limit to %d\n", ident->locallimit);
369 printlog(LOG_DEBUG, "%d Limit ID:%d\n", ident->locallimit, ident->id);
371 snprintf(cmd, CMD_BUFFER_SIZE,
372 "/sbin/tc class change dev eth0 parent 1:%x classid 1:%x htb rate 8bit ceil %dbps quantum 1600",
373 ident->htb_parent, ident->htb_node, ident->locallimit);
378 /* FIXME: call failed. What to do? */
383 for (i = 0; i < ident->leaf_count; ++i) {
384 if (ident->drop_prob >= ident->leaves[i]->drop_prob) {
385 /* The new drop probability for this identity is greater
386 * than or equal to the leaf's current drop probability.
387 * We can safely use the larger value at this leaf
389 ident->leaves[i]->drop_prob = ident->drop_prob;
390 } else if (ident->last_drop_prob < ident->leaves[i]->drop_prob) {
391 /* The old drop probability for this identity is less than
392 * the leaf's current drop probability. This means that
393 * this identity couldn't have been the limiting ident,
394 * so nothing needs to be done because the old limiting
395 * ident is still the limiting factor. */
397 /* Intentionally blank. */
399 /* If neither of the above are true, then...
400 * 1) The new drop probability for the identity is less
401 * than what it previously was, and
402 * 2) This ident may have had the maximum drop probability
403 * of all idents limiting this leaf, and therefore we need
404 * to follow the leaf's parents up to the root to find the
405 * new leaf drop probability safely. */
406 ident->leaves[i]->drop_prob =
407 find_leaf_drop_prob(ident->leaves[i]);
410 /* Make the call to tc. */
412 snprintf(cmd, CMD_BUFFER_SIZE,
413 "/sbin/tc qdisc change dev eth0 parent 1:1%x handle 1%x netem loss %.4f delay 40ms",
414 ident->leaves[i]->xid, ident->leaves[i]->xid,
415 (100 * ident->leaves[i]->drop_prob));
417 snprintf(cmd, CMD_BUFFER_SIZE,
418 "/sbin/tc qdisc change dev eth0 parent 1:1%x handle 1%x netem loss %.4f delay 0ms",
419 ident->leaves[i]->xid, ident->leaves[i]->xid,
420 (100 * ident->leaves[i]->drop_prob));
425 /* FIXME: call failed. What to do? */
432 printlog(LOG_CRITICAL, "DRL enforce: unknown policy %d\n",limiter->policy);
440 * This function is periodically called to clean the stable instance's flow
441 * accounting tables for each identity.
443 static void clean(drl_instance_t *instance) {
444 identity_t *ident = NULL;
446 map_reset_iterate(instance->ident_map);
447 while ((ident = map_next(instance->ident_map)) != NULL) {
448 pthread_mutex_lock(&ident->table_mutex);
450 ident->table_cleanup_function(ident->table);
452 pthread_mutex_unlock(&ident->table_mutex);
455 /* Periodically flush the log file. */
459 static void print_averages(drl_instance_t *instance, int print_interval) {
460 identity_t *ident = NULL;
462 map_reset_iterate(instance->ident_map);
463 while ((ident = map_next(instance->ident_map)) != NULL) {
464 ident->avg_bytes /= (double) print_interval;
465 //printf("avg_bytes = %f, print_interval = %d\n", ident->avg_bytes, print_interval);
466 printlog(LOG_DEBUG, "%.3f \t Avg rate. ID:%d\n",
467 ident->avg_bytes / 128, ident->id);
468 //printf("%.3f \t Avg rate. ID:%d\n",
469 // ident->avg_bytes / 128, ident->id);
470 ident->avg_bytes = 0;
474 /** Thread function to handle local rate estimation.
476 * None of our simple hashmap functions are thread safe, so we lock the limiter
477 * with an rwlock to prevent another thread from attempting to modify the set
480 * Each identity also has a private lock for its table. This gets locked by
481 * table-modifying functions such as estimate and clean.
483 void handle_estimation(void *arg) {
484 limiter_t *limiter = (limiter_t *) arg;
485 identity_t *ident = NULL;
486 int clean_timer, clean_wait_intervals;
487 useconds_t sleep_time = limiter->estintms * 1000;
488 uint32_t cal_slot = 0;
489 int print_interval = 1000 / (limiter->estintms);
491 sigset_t signal_mask;
493 sigemptyset(&signal_mask);
494 sigaddset(&signal_mask, SIGHUP);
495 pthread_sigmask(SIG_BLOCK, &signal_mask, NULL);
497 /* Determine the number of intervals we should wait before hitting the
498 * specified clean interval. (Converts seconds -> intervals). */
499 clean_wait_intervals = IDENT_CLEAN_INTERVAL * (1000.0 / limiter->estintms);
500 clean_timer = clean_wait_intervals;
503 /* Sleep according to the delay of the estimate interval. */
506 /* Grab the limiter lock for reading. This prevents identities from
507 * disappearing beneath our feet. */
508 pthread_rwlock_rdlock(&limiter->limiter_lock);
510 cal_slot = limiter->stable_instance.cal_slot & SCHEDMASK;
512 /* Service all the identities that are scheduled to run during this
514 while (!TAILQ_EMPTY(limiter->stable_instance.cal + cal_slot)) {
515 ident = TAILQ_FIRST(limiter->stable_instance.cal + cal_slot);
516 TAILQ_REMOVE(limiter->stable_instance.cal + cal_slot, ident, calendar);
518 /* Update the ident's flow accouting table with the latest info. */
521 /* Determine its share of the rate allocation. */
522 allocate(limiter, ident);
524 /* Make tc calls to enforce the rate we decided upon. */
525 enforce(limiter, ident);
527 /* Tell the comm library to propagate this identity's result for
529 send_update(&ident->comm, ident->id);
531 /* Add ident back to the queue at a future time slot. */
532 TAILQ_INSERT_TAIL(limiter->stable_instance.cal +
533 ((cal_slot + ident->intervals) & SCHEDMASK),
538 if (loglevel() == LOG_DEBUG && print_interval <= 0) {
539 print_interval = 1000 / (limiter->estintms);
540 print_averages(&limiter->stable_instance, print_interval);
543 /* Check if enough intervals have passed for cleaning. */
544 if (clean_timer <= 0) {
545 clean(&limiter->stable_instance);
546 clean_timer = clean_wait_intervals;
551 limiter->stable_instance.cal_slot += 1;
553 pthread_rwlock_unlock(&limiter->limiter_lock);