1 /* Copyright (c) 2004-2017. The SimGrid Team. 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 #ifndef SURF_MAXMIN_HPP
7 #define SURF_MAXMIN_HPP
9 #include "src/internal_config.h"
10 #include "surf/surf.hpp"
11 #include "xbt/asserts.h"
15 /** @addtogroup SURF_lmm
17 * A linear maxmin solver to resolve inequations systems.
19 * Most SimGrid model rely on a "fluid/steady-state" modeling that simulate the sharing of resources between actions at
20 * relatively coarse-grain. Such sharing is generally done by solving a set of linear inequations. Let's take an
21 * example and assume we have the variables \f$x_1\f$, \f$x_2\f$, \f$x_3\f$, and \f$x_4\f$ . Let's say that \f$x_1\f$
22 * and \f$x_2\f$ correspond to activities running and the same CPU \f$A\f$ whose capacity is \f$C_A\f$. In such a
23 * case, we need to enforce:
25 * \f[ x_1 + x_2 \leq C_A \f]
27 * Likewise, if \f$x_3\f$ (resp. \f$x_4\f$) corresponds to a network flow \f$F_3\f$ (resp. \f$F_4\f$) that goes through
28 * a set of links \f$L_1\f$ and \f$L_2\f$ (resp. \f$L_2\f$ and \f$L_3\f$), then we need to enforce:
30 * \f[ x_3 \leq C_{L_1} \f]
31 * \f[ x_3 + x_4 \leq C_{L_2} \f]
32 * \f[ x_4 \leq C_{L_3} \f]
34 * One could set every variable to 0 to make sure the constraints are satisfied but this would obviously not be very
35 * realistic. A possible objective is to try to maximize the minimum of the \f$x_i\f$ . This ensures that all the
36 * \f$x_i\f$ are positive and "as large as possible".
38 * This is called *max-min fairness* and is the most commonly used objective in SimGrid. Another possibility is to
39 * maximize \f$\sum_if(x_i)\f$, where \f$f\f$ is a strictly increasing concave function.
54 * A possible system could be:
55 * - three variables: `var1`, `var2`, `var3`
56 * - two constraints: `cons1`, `cons2`
57 * - four elements linking:
58 * - `elem1` linking `var1` and `cons1`
59 * - `elem2` linking `var2` and `cons1`
60 * - `elem3` linking `var2` and `cons2`
61 * - `elem4` linking `var3` and `cons2`
63 * And the corresponding inequations will be:
65 * var1.value <= var1.bound
66 * var2.value <= var2.bound
67 * var3.value <= var3.bound
68 * var1.weight * var1.value * elem1.value + var2.weight * var2.value * elem2.value <= cons1.bound
69 * var2.weight * var2.value * elem3.value + var3.weight * var3.value * elem4.value <= cons2.bound
71 * where `var1.value`, `var2.value` and `var3.value` are the unknown values.
73 * If a constraint is not shared, the sum is replaced by a max.
74 * For example, a third non-shared constraint `cons3` and the associated elements `elem5` and `elem6` could write as:
76 * max( var1.weight * var1.value * elem5.value , var3.weight * var3.value * elem6.value ) <= cons3.bound
78 * This is usefull for the sharing of resources for various models.
79 * For instance, for the network model, each link is associated to a constraint and each communication to a variable.
81 * Implementation details
83 * For implementation reasons, we are interested in distinguishing variables that actually participate to the
84 * computation of constraints, and those who are part of the equations but are stuck to zero.
85 * We call enabled variables, those which var.weight is strictly positive. Zero-weight variables are called disabled
87 * Unfortunately this concept of enabled/disabled variables intersects with active/inactive variable.
88 * Semantically, the intent is similar, but the conditions under which a variable is active is slightly more strict
89 * than the conditions for it to be enabled.
90 * A variable is active only if its var.value is non-zero (and, by construction, its var.weight is non-zero).
91 * In general, variables remain disabled after their creation, which often models an initialization phase (e.g. first
92 * packet propagating in the network). Then, it is enabled by the corresponding model. Afterwards, the max-min solver
93 * (lmm_solve()) activates it when appropriate. It is possible that the variable is again disabled, e.g. to model the
94 * pausing of an action.
96 * Concurrency limit and maximum
98 * We call concurrency, the number of variables that can be enabled at any time for each constraint.
99 * From a model perspective, this "concurrency" often represents the number of actions that actually compete for one
101 * The LMM solver is able to limit the concurrency for each constraint, and to monitor its maximum value.
103 * One may want to limit the concurrency of constraints for essentially three reasons:
104 * - Keep LMM system in a size that can be solved (it does not react very well with tens of thousands of variables per
106 * - Stay within parameters where the fluid model is accurate enough.
107 * - Model serialization effects
109 * The concurrency limit can also be set to a negative value to disable concurrency limit. This can improve performance
112 * Overall, each constraint contains three fields related to concurrency:
113 * - concurrency_limit which is the limit enforced by the solver
114 * - concurrency_current which is the current concurrency
115 * - concurrency_maximum which is the observed maximum concurrency
117 * Variables also have one field related to concurrency: concurrency_share.
118 * In effect, in some cases, one variable is involved multiple times (i.e. two elements) in a constraint.
119 * For example, cross-traffic is modeled using 2 elements per constraint.
120 * concurrency_share formally corresponds to the maximum number of elements that associate the variable and any given
124 XBT_PUBLIC_DATA(double) sg_maxmin_precision;
125 XBT_PUBLIC_DATA(double) sg_surf_precision;
126 XBT_PUBLIC_DATA(int) sg_concurrency_limit;
128 static inline void double_update(double* variable, double value, double precision)
130 // printf("Updating %g -= %g +- %g\n",*variable,value,precision);
131 // xbt_assert(value==0 || value>precision);
132 // Check that precision is higher than the machine-dependent size of the mantissa. If not, brutal rounding may
133 // happen, and the precision mechanism is not active...
134 // xbt_assert(*variable< (2<<DBL_MANT_DIG)*precision && FLT_RADIX==2);
136 if (*variable < precision)
140 static inline int double_positive(double value, double precision)
142 return (value > precision);
145 static inline int double_equals(double value1, double value2, double precision)
147 return (fabs(value1 - value2) < precision);
150 /** @{ @ingroup SURF_lmm */
153 * @brief Get the value of the variable after the last lmm solve
154 * @param var A variable
155 * @return The value of the variable
157 XBT_PUBLIC(double) lmm_variable_getvalue(lmm_variable_t var);
160 * @brief Get the maximum value of the variable (-1.0 if no maximum value)
161 * @param var A variable
162 * @return The bound of the variable
164 XBT_PUBLIC(double) lmm_variable_getbound(lmm_variable_t var);
167 * @brief Set the concurrent share of the variable
168 * @param var A variable
169 * @param concurrency_share The new concurrency share
171 XBT_PUBLIC(void) lmm_variable_concurrency_share_set(lmm_variable_t var, short int concurrency_share);
174 * @brief Get the numth constraint associated to the variable
175 * @param sys The system associated to the variable (not used)
176 * @param var A variable
177 * @param num The rank of constraint we want to get
178 * @return The numth constraint
180 XBT_PUBLIC(lmm_constraint_t) lmm_get_cnst_from_var(lmm_system_t sys, lmm_variable_t var, unsigned num);
183 * @brief Get the weigth of the numth constraint associated to the variable
184 * @param sys The system associated to the variable (not used)
185 * @param var A variable
186 * @param num The rank of constraint we want to get
187 * @return The numth constraint
189 XBT_PUBLIC(double) lmm_get_cnst_weight_from_var(lmm_system_t sys, lmm_variable_t var, unsigned num);
192 * @brief Get the number of constraint associated to a variable
193 * @param sys The system associated to the variable (not used)
194 * @param var A variable
195 * @return The number of constraint associated to the variable
197 XBT_PUBLIC(int) lmm_get_number_of_cnst_from_var(lmm_system_t sys, lmm_variable_t var);
200 * @brief Get the data associated to a variable
201 * @param var A variable
202 * @return The data associated to the variable
204 XBT_PUBLIC(void*) lmm_variable_id(lmm_variable_t var);
207 * @brief Get the weight of a variable
208 * @param var A variable
209 * @return The weight of the variable
211 XBT_PUBLIC(double) lmm_get_variable_weight(lmm_variable_t var);
214 * @brief Solve the lmm system
215 * @param sys The lmm system to solve
217 XBT_PUBLIC(void) lmm_solve(lmm_system_t sys);
219 XBT_PUBLIC(void) lagrange_solve(lmm_system_t sys);
220 XBT_PUBLIC(void) bottleneck_solve(lmm_system_t sys);
222 /** Default functions associated to the chosen protocol. When using the lagrangian approach. */
225 lmm_set_default_protocol_function(double (*func_f)(lmm_variable_t var, double x),
226 double (*func_fp)(lmm_variable_t var, double x),
227 double (*func_fpi)(lmm_variable_t var, double x));
229 XBT_PUBLIC(double) func_reno_f(lmm_variable_t var, double x);
230 XBT_PUBLIC(double) func_reno_fp(lmm_variable_t var, double x);
231 XBT_PUBLIC(double) func_reno_fpi(lmm_variable_t var, double x);
233 XBT_PUBLIC(double) func_reno2_f(lmm_variable_t var, double x);
234 XBT_PUBLIC(double) func_reno2_fp(lmm_variable_t var, double x);
235 XBT_PUBLIC(double) func_reno2_fpi(lmm_variable_t var, double x);
237 XBT_PUBLIC(double) func_vegas_f(lmm_variable_t var, double x);
238 XBT_PUBLIC(double) func_vegas_fp(lmm_variable_t var, double x);
239 XBT_PUBLIC(double) func_vegas_fpi(lmm_variable_t var, double x);