2 /* Copyright (c) 2007 Arnaud Legrand, Pedro Velho. All rights reserved. */
3 /* This program is free software; you can redistribute it and/or modify it
4 * under the terms of the license (GNU LGPL) which comes with this package. */
6 * Modelling the proportional fairness using the Lagrange Optimization
7 * Approach. For a detailed description see:
8 * "ssh://username@scm.gforge.inria.fr/svn/memo/people/pvelho/lagrange/ppf.ps".
11 #include "xbt/sysdep.h"
12 #include "xbt/mallocator.h"
13 #include "maxmin_private.h"
20 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_lagrange, surf,
21 "Logging specific to SURF (lagrange)");
22 XBT_LOG_NEW_SUBCATEGORY(surf_lagrange_dichotomy, surf_lagrange,
23 "Logging specific to SURF (lagrange dichotomy)");
25 #define SHOW_EXPR(expr) CDEBUG1(surf_lagrange,#expr " = %g",expr);
27 double (*func_f_def) (lmm_variable_t, double);
28 double (*func_fp_def) (lmm_variable_t, double);
29 double (*func_fpi_def) (lmm_variable_t, double);
32 * Local prototypes to implement the lagrangian optimization with optimal step, also called dichotomy.
34 //solves the proportional fairness using a lagrange optimizition with dichotomy step
35 void lagrange_solve(lmm_system_t sys);
36 //computes the value of the dichotomy using a initial values, init, with a specific variable or constraint
37 static double dichotomy(double init, double diff(double, void *),
38 void *var_cnst, double min_error);
39 //computes the value of the differential of variable param_var applied to mu
40 static double partial_diff_mu(double mu, void *param_var);
41 //computes the value of the differential of constraint param_cnst applied to lambda
42 static double partial_diff_lambda(double lambda, void *param_cnst);
44 static int __check_feasible(xbt_swag_t cnst_list, xbt_swag_t var_list,
47 xbt_swag_t elem_list = NULL;
48 lmm_element_t elem = NULL;
49 lmm_constraint_t cnst = NULL;
50 lmm_variable_t var = NULL;
54 xbt_swag_foreach(cnst, cnst_list) {
56 elem_list = &(cnst->element_set);
57 xbt_swag_foreach(elem, elem_list) {
64 if (double_positive(tmp - cnst->bound)) {
67 ("The link (%p) is over-used. Expected less than %f and got %f",
68 cnst, cnst->bound, tmp);
72 ("Checking feasability for constraint (%p): sat = %f, lambda = %f ",
73 cnst, tmp - cnst->bound, cnst->lambda);
76 xbt_swag_foreach(var, var_list) {
81 DEBUG3("Checking feasability for variable (%p): sat = %f mu = %f", var,
82 var->value - var->bound, var->mu);
84 if (double_positive(var->value - var->bound)) {
87 ("The variable (%p) is too large. Expected less than %f and got %f",
88 var, var->bound, var->value);
95 static double new_value(lmm_variable_t var)
100 for (i = 0; i < var->cnsts_number; i++) {
101 tmp += (var->cnsts[i].constraint)->lambda;
105 DEBUG3("\t Working on var (%p). cost = %e; Df = %e", var, tmp, var->df);
106 //uses the partial differential inverse function
107 return var->func_fpi(var, tmp);
110 static double new_mu(lmm_variable_t var)
113 double sigma_i = 0.0;
116 for (j = 0; j < var->cnsts_number; j++) {
117 sigma_i += (var->cnsts[j].constraint)->lambda;
119 mu_i = var->func_fp(var, var->bound) - sigma_i;
125 static double dual_objective(xbt_swag_t var_list, xbt_swag_t cnst_list)
127 lmm_constraint_t cnst = NULL;
128 lmm_variable_t var = NULL;
132 xbt_swag_foreach(var, var_list) {
133 double sigma_i = 0.0;
139 for (j = 0; j < var->cnsts_number; j++)
140 sigma_i += (var->cnsts[j].constraint)->lambda;
145 DEBUG2("var %p : sigma_i = %1.20f", var, sigma_i);
147 obj += var->func_f(var, var->func_fpi(var, sigma_i)) -
148 sigma_i * var->func_fpi(var, sigma_i);
151 obj += var->mu * var->bound;
154 xbt_swag_foreach(cnst, cnst_list)
155 obj += cnst->lambda * cnst->bound;
160 void lagrange_solve(lmm_system_t sys)
163 * Lagrange Variables.
165 int max_iterations = 100;
166 double epsilon_min_error = MAXMIN_PRECISION;
167 double dichotomy_min_error = 1e-14;
168 double overall_modification = 1;
171 * Variables to manipulate the data structure proposed to model the maxmin
172 * fairness. See docummentation for more details.
174 xbt_swag_t cnst_list = NULL;
175 lmm_constraint_t cnst = NULL;
177 xbt_swag_t var_list = NULL;
178 lmm_variable_t var = NULL;
181 * Auxiliar variables.
188 DEBUG0("Iterative method configuration snapshot =====>");
189 DEBUG1("#### Maximum number of iterations : %d", max_iterations);
190 DEBUG1("#### Minimum error tolerated : %e",
192 DEBUG1("#### Minimum error tolerated (dichotomy) : %e",
193 dichotomy_min_error);
195 if (XBT_LOG_ISENABLED(surf_lagrange, xbt_log_priority_debug)) {
199 if (!(sys->modified))
206 cnst_list = &(sys->active_constraint_set);
207 xbt_swag_foreach(cnst, cnst_list) {
209 cnst->new_lambda = 2.0;
210 DEBUG2("#### cnst(%p)->lambda : %e", cnst, cnst->lambda);
214 * Initialize the var list variable with only the active variables.
215 * Associate an index in the swag variables. Initialize mu.
217 var_list = &(sys->variable_set);
219 xbt_swag_foreach(var, var_list) {
224 if (var->bound < 0.0) {
225 DEBUG1("#### NOTE var(%d) is a boundless variable", i);
227 var->value = new_value(var);
231 var->value = new_value(var);
233 DEBUG2("#### var(%p) ->df : %e", var, var->df);
234 DEBUG2("#### var(%p) ->mu : %e", var, var->mu);
235 DEBUG2("#### var(%p) ->weight: %e", var, var->weight);
236 DEBUG2("#### var(%p) ->bound: %e", var, var->bound);
237 for (i = 0; i < var->cnsts_number; i++) {
238 if(var->cnsts[i].value==0.0) nb++;
240 if(nb==var->cnsts_number) var->value = 1.0;
245 * Compute dual objective.
247 obj = dual_objective(var_list, cnst_list);
250 * While doesn't reach a minimun error or a number maximum of iterations.
252 while (overall_modification > epsilon_min_error
253 && iteration < max_iterations) {
254 /* int dual_updated=0; */
257 DEBUG1("************** ITERATION %d **************", iteration);
258 DEBUG0("-------------- Gradient Descent ----------");
261 * Improve the value of mu_i
263 xbt_swag_foreach(var, var_list) {
266 if (var->bound >= 0) {
267 DEBUG1("Working on var (%p)", var);
268 var->new_mu = new_mu(var);
269 /* dual_updated += (fabs(var->new_mu-var->mu)>dichotomy_min_error); */
270 /* DEBUG2("dual_updated (%d) : %1.20f",dual_updated,fabs(var->new_mu-var->mu)); */
271 DEBUG3("Updating mu : var->mu (%p) : %1.20f -> %1.20f", var,
272 var->mu, var->new_mu);
273 var->mu = var->new_mu;
275 new_obj = dual_objective(var_list, cnst_list);
276 DEBUG3("Improvement for Objective (%g -> %g) : %g", obj, new_obj,
278 xbt_assert1(obj - new_obj >= -epsilon_min_error,
279 "Our gradient sucks! (%1.20f)", obj - new_obj);
285 * Improve the value of lambda_i
287 xbt_swag_foreach(cnst, cnst_list) {
288 DEBUG1("Working on cnst (%p)", cnst);
290 dichotomy(cnst->lambda, partial_diff_lambda, cnst,
291 dichotomy_min_error);
292 /* dual_updated += (fabs(cnst->new_lambda-cnst->lambda)>dichotomy_min_error); */
293 /* DEBUG2("dual_updated (%d) : %1.20f",dual_updated,fabs(cnst->new_lambda-cnst->lambda)); */
294 DEBUG3("Updating lambda : cnst->lambda (%p) : %1.20f -> %1.20f",
295 cnst, cnst->lambda, cnst->new_lambda);
296 cnst->lambda = cnst->new_lambda;
298 new_obj = dual_objective(var_list, cnst_list);
299 DEBUG3("Improvement for Objective (%g -> %g) : %g", obj, new_obj,
301 xbt_assert1(obj - new_obj >= -epsilon_min_error,
302 "Our gradient sucks! (%1.20f)", obj - new_obj);
307 * Now computes the values of each variable (\rho) based on
308 * the values of \lambda and \mu.
310 DEBUG0("-------------- Check convergence ----------");
311 overall_modification = 0;
312 xbt_swag_foreach(var, var_list) {
313 if (var->weight <= 0)
316 tmp = new_value(var);
318 overall_modification =
319 MAX(overall_modification, fabs(var->value - tmp));
322 DEBUG3("New value of var (%p) = %e, overall_modification = %e",
323 var, var->value, overall_modification);
327 DEBUG0("-------------- Check feasability ----------");
328 if (!__check_feasible(cnst_list, var_list, 0))
329 overall_modification = 1.0;
330 DEBUG2("Iteration %d: overall_modification : %f", iteration,
331 overall_modification);
332 /* if(!dual_updated) { */
333 /* WARN1("Could not improve the convergence at iteration %d. Drop it!",iteration); */
338 __check_feasible(cnst_list, var_list, 1);
340 if (overall_modification <= epsilon_min_error) {
341 DEBUG1("The method converges in %d iterations.", iteration);
343 if (iteration >= max_iterations) {
345 ("Method reach %d iterations, which is the maximum number of iterations allowed.",
348 /* INFO1("Method converged after %d iterations", iteration); */
350 if (XBT_LOG_ISENABLED(surf_lagrange, xbt_log_priority_debug)) {
356 * Returns a double value corresponding to the result of a dichotomy proccess with
357 * respect to a given variable/constraint (\mu in the case of a variable or \lambda in
358 * case of a constraint) and a initial value init.
360 * @param init initial value for \mu or \lambda
361 * @param diff a function that computes the differential of with respect a \mu or \lambda
362 * @param var_cnst a pointer to a variable or constraint
363 * @param min_erro a minimun error tolerated
365 * @return a double correponding to the result of the dichotomyal process
367 static double dichotomy(double init, double diff(double, void *),
368 void *var_cnst, double min_error)
371 double overall_error;
373 double min_diff, max_diff, middle_diff;
383 min_diff = max_diff = middle_diff = 0.0;
386 if ((diff_0 = diff(1e-16, var_cnst)) >= 0) {
387 CDEBUG1(surf_lagrange_dichotomy, "returning 0.0 (diff = %e)", diff_0);
392 min_diff = diff(min, var_cnst);
393 max_diff = diff(max, var_cnst);
395 while (overall_error > min_error) {
396 CDEBUG4(surf_lagrange_dichotomy,
397 "[min, max] = [%1.20f, %1.20f] || diffmin, diffmax = %1.20f, %1.20f",
398 min, max, min_diff, max_diff);
400 if (min_diff > 0 && max_diff > 0) {
402 CDEBUG0(surf_lagrange_dichotomy, "Decreasing min");
404 min_diff = diff(min, var_cnst);
406 CDEBUG0(surf_lagrange_dichotomy, "Decreasing max");
410 } else if (min_diff < 0 && max_diff < 0) {
412 CDEBUG0(surf_lagrange_dichotomy, "Increasing max");
414 max_diff = diff(max, var_cnst);
416 CDEBUG0(surf_lagrange_dichotomy, "Increasing min");
420 } else if (min_diff < 0 && max_diff > 0) {
421 middle = (max + min) / 2.0;
422 CDEBUG1(surf_lagrange_dichotomy, "Trying (max+min)/2 : %1.20f",
425 if ((min == middle) || (max == middle)) {
426 CWARN4(surf_lagrange_dichotomy,
427 "Cannot improve the convergence! min=max=middle=%1.20f, diff = %1.20f."
428 " Reaching the 'double' limits. Maybe scaling your function would help ([%1.20f,%1.20f]).",
429 min, max - min, min_diff, max_diff);
432 middle_diff = diff(middle, var_cnst);
434 if (middle_diff < 0) {
435 CDEBUG0(surf_lagrange_dichotomy, "Increasing min");
437 overall_error = max_diff - middle_diff;
438 min_diff = middle_diff;
439 /* SHOW_EXPR(overall_error); */
440 } else if (middle_diff > 0) {
441 CDEBUG0(surf_lagrange_dichotomy, "Decreasing max");
443 overall_error = max_diff - middle_diff;
444 max_diff = middle_diff;
445 /* SHOW_EXPR(overall_error); */
448 /* SHOW_EXPR(overall_error); */
450 } else if (min_diff == 0) {
453 /* SHOW_EXPR(overall_error); */
454 } else if (max_diff == 0) {
457 /* SHOW_EXPR(overall_error); */
458 } else if (min_diff > 0 && max_diff < 0) {
459 CWARN0(surf_lagrange_dichotomy,
460 "The impossible happened, partial_diff(min) > 0 && partial_diff(max) < 0");
463 CWARN2(surf_lagrange_dichotomy,
464 "diffmin (%1.20f) or diffmax (%1.20f) are something I don't know, taking no action.",
470 CDEBUG1(surf_lagrange_dichotomy, "returning %e", (min + max) / 2.0);
472 return ((min + max) / 2.0);
475 static double partial_diff_lambda(double lambda, void *param_cnst)
479 xbt_swag_t elem_list = NULL;
480 lmm_element_t elem = NULL;
481 lmm_variable_t var = NULL;
482 lmm_constraint_t cnst = (lmm_constraint_t) param_cnst;
484 double sigma_i = 0.0;
487 elem_list = &(cnst->element_set);
489 CDEBUG1(surf_lagrange_dichotomy, "Computing diff of cnst (%p)", cnst);
491 xbt_swag_foreach(elem, elem_list) {
492 var = elem->variable;
493 if (var->weight <= 0)
496 CDEBUG1(surf_lagrange_dichotomy, "Computing sigma_i for var (%p)",
498 // Initialize the summation variable
502 for (j = 0; j < var->cnsts_number; j++) {
503 sigma_i += (var->cnsts[j].constraint)->lambda;
506 //add mu_i if this flow has a RTT constraint associated
510 //replace value of cnst->lambda by the value of parameter lambda
511 sigma_i = (sigma_i - cnst->lambda) + lambda;
513 diff += -var->func_fpi(var, sigma_i);
519 CDEBUG3(surf_lagrange_dichotomy,
520 "d D/d lambda for cnst (%p) at %1.20f = %1.20f", cnst, lambda,
526 /** \brief Attribute the value bound to var->bound.
528 * \param func_fpi inverse of the partial differential of f (f prime inverse, (f')^{-1})
530 * Set default functions to the ones passed as parameters. This is a polimorfism in C pure, enjoy the roots of programming.
534 lmm_set_default_protocol_function(double (*func_f)
535 (lmm_variable_t var, double x),
536 double (*func_fp) (lmm_variable_t var,
538 double (*func_fpi) (lmm_variable_t var,
542 func_fp_def = func_fp;
543 func_fpi_def = func_fpi;
547 /**************** Vegas and Reno functions *************************/
549 * NOTE for Reno: all functions consider the network
550 * coeficient (alpha) equal to 1.
554 * For Vegas: $f(x) = \alpha D_f\ln(x)$
555 * Therefore: $fp(x) = \frac{\alpha D_f}{x}$
556 * Therefore: $fpi(x) = \frac{\alpha D_f}{x}$
558 #define VEGAS_SCALING 1000.0
560 double func_vegas_f(lmm_variable_t var, double x)
562 xbt_assert1(x > 0.0, "Don't call me with stupid values! (%1.20f)", x);
563 return VEGAS_SCALING * var->df * log(x);
566 double func_vegas_fp(lmm_variable_t var, double x)
568 xbt_assert1(x > 0.0, "Don't call me with stupid values! (%1.20f)", x);
569 return VEGAS_SCALING * var->df / x;
572 double func_vegas_fpi(lmm_variable_t var, double x)
574 xbt_assert1(x > 0.0, "Don't call me with stupid values! (%1.20f)", x);
575 return var->df / (x / VEGAS_SCALING);
579 * For Reno: $f(x) = \frac{\sqrt{3/2}}{D_f} atan(\sqrt{3/2}D_f x)$
580 * Therefore: $fp(x) = \frac{3}{3 D_f^2 x^2+2}$
581 * Therefore: $fpi(x) = \sqrt{\frac{1}{{D_f}^2 x} - \frac{2}{3{D_f}^2}}$
583 #define RENO_SCALING 1.0
584 double func_reno_f(lmm_variable_t var, double x)
586 xbt_assert0(var->df > 0.0, "Don't call me with stupid values!");
588 return RENO_SCALING * sqrt(3.0 / 2.0) / var->df * atan(sqrt(3.0 / 2.0) *
592 double func_reno_fp(lmm_variable_t var, double x)
594 return RENO_SCALING * 3.0 / (3.0 * var->df * var->df * x * x + 2.0);
597 double func_reno_fpi(lmm_variable_t var, double x)
601 xbt_assert0(var->df > 0.0, "Don't call me with stupid values!");
602 xbt_assert0(x > 0.0, "Don't call me with stupid values!");
605 1.0 / (var->df * var->df * (x / RENO_SCALING)) -
606 2.0 / (3.0 * var->df * var->df);
609 /* xbt_assert0(res_fpi>0.0,"Don't call me with stupid values!"); */
610 return sqrt(res_fpi);