hugetlb: support boot allocate different sizes
[linux-2.6/next.git] / include / net / red.h
blob3cf31d466a81254e019aac768666458ffbae6009
1 #ifndef __NET_SCHED_RED_H
2 #define __NET_SCHED_RED_H
4 #include <linux/types.h>
5 #include <net/pkt_sched.h>
6 #include <net/inet_ecn.h>
7 #include <net/dsfield.h>
9 /* Random Early Detection (RED) algorithm.
10 =======================================
12 Source: Sally Floyd and Van Jacobson, "Random Early Detection Gateways
13 for Congestion Avoidance", 1993, IEEE/ACM Transactions on Networking.
15 This file codes a "divisionless" version of RED algorithm
16 as written down in Fig.17 of the paper.
18 Short description.
19 ------------------
21 When a new packet arrives we calculate the average queue length:
23 avg = (1-W)*avg + W*current_queue_len,
25 W is the filter time constant (chosen as 2^(-Wlog)), it controls
26 the inertia of the algorithm. To allow larger bursts, W should be
27 decreased.
29 if (avg > th_max) -> packet marked (dropped).
30 if (avg < th_min) -> packet passes.
31 if (th_min < avg < th_max) we calculate probability:
33 Pb = max_P * (avg - th_min)/(th_max-th_min)
35 and mark (drop) packet with this probability.
36 Pb changes from 0 (at avg==th_min) to max_P (avg==th_max).
37 max_P should be small (not 1), usually 0.01..0.02 is good value.
39 max_P is chosen as a number, so that max_P/(th_max-th_min)
40 is a negative power of two in order arithmetics to contain
41 only shifts.
44 Parameters, settable by user:
45 -----------------------------
47 qth_min - bytes (should be < qth_max/2)
48 qth_max - bytes (should be at least 2*qth_min and less limit)
49 Wlog - bits (<32) log(1/W).
50 Plog - bits (<32)
52 Plog is related to max_P by formula:
54 max_P = (qth_max-qth_min)/2^Plog;
56 F.e. if qth_max=128K and qth_min=32K, then Plog=22
57 corresponds to max_P=0.02
59 Scell_log
60 Stab
62 Lookup table for log((1-W)^(t/t_ave).
65 NOTES:
67 Upper bound on W.
68 -----------------
70 If you want to allow bursts of L packets of size S,
71 you should choose W:
73 L + 1 - th_min/S < (1-(1-W)^L)/W
75 th_min/S = 32 th_min/S = 4
77 log(W) L
78 -1 33
79 -2 35
80 -3 39
81 -4 46
82 -5 57
83 -6 75
84 -7 101
85 -8 135
86 -9 190
87 etc.
90 #define RED_STAB_SIZE 256
91 #define RED_STAB_MASK (RED_STAB_SIZE - 1)
93 struct red_stats
95 u32 prob_drop; /* Early probability drops */
96 u32 prob_mark; /* Early probability marks */
97 u32 forced_drop; /* Forced drops, qavg > max_thresh */
98 u32 forced_mark; /* Forced marks, qavg > max_thresh */
99 u32 pdrop; /* Drops due to queue limits */
100 u32 other; /* Drops due to drop() calls */
101 u32 backlog;
104 struct red_parms
106 /* Parameters */
107 u32 qth_min; /* Min avg length threshold: A scaled */
108 u32 qth_max; /* Max avg length threshold: A scaled */
109 u32 Scell_max;
110 u32 Rmask; /* Cached random mask, see red_rmask */
111 u8 Scell_log;
112 u8 Wlog; /* log(W) */
113 u8 Plog; /* random number bits */
114 u8 Stab[RED_STAB_SIZE];
116 /* Variables */
117 int qcount; /* Number of packets since last random
118 number generation */
119 u32 qR; /* Cached random number */
121 unsigned long qavg; /* Average queue length: A scaled */
122 psched_time_t qidlestart; /* Start of current idle period */
125 static inline u32 red_rmask(u8 Plog)
127 return Plog < 32 ? ((1 << Plog) - 1) : ~0UL;
130 static inline void red_set_parms(struct red_parms *p,
131 u32 qth_min, u32 qth_max, u8 Wlog, u8 Plog,
132 u8 Scell_log, u8 *stab)
134 /* Reset average queue length, the value is strictly bound
135 * to the parameters below, reseting hurts a bit but leaving
136 * it might result in an unreasonable qavg for a while. --TGR
138 p->qavg = 0;
140 p->qcount = -1;
141 p->qth_min = qth_min << Wlog;
142 p->qth_max = qth_max << Wlog;
143 p->Wlog = Wlog;
144 p->Plog = Plog;
145 p->Rmask = red_rmask(Plog);
146 p->Scell_log = Scell_log;
147 p->Scell_max = (255 << Scell_log);
149 memcpy(p->Stab, stab, sizeof(p->Stab));
152 static inline int red_is_idling(struct red_parms *p)
154 return p->qidlestart != PSCHED_PASTPERFECT;
157 static inline void red_start_of_idle_period(struct red_parms *p)
159 p->qidlestart = psched_get_time();
162 static inline void red_end_of_idle_period(struct red_parms *p)
164 p->qidlestart = PSCHED_PASTPERFECT;
167 static inline void red_restart(struct red_parms *p)
169 red_end_of_idle_period(p);
170 p->qavg = 0;
171 p->qcount = -1;
174 static inline unsigned long red_calc_qavg_from_idle_time(struct red_parms *p)
176 psched_time_t now;
177 long us_idle;
178 int shift;
180 now = psched_get_time();
181 us_idle = psched_tdiff_bounded(now, p->qidlestart, p->Scell_max);
184 * The problem: ideally, average length queue recalcultion should
185 * be done over constant clock intervals. This is too expensive, so
186 * that the calculation is driven by outgoing packets.
187 * When the queue is idle we have to model this clock by hand.
189 * SF+VJ proposed to "generate":
191 * m = idletime / (average_pkt_size / bandwidth)
193 * dummy packets as a burst after idle time, i.e.
195 * p->qavg *= (1-W)^m
197 * This is an apparently overcomplicated solution (f.e. we have to
198 * precompute a table to make this calculation in reasonable time)
199 * I believe that a simpler model may be used here,
200 * but it is field for experiments.
203 shift = p->Stab[(us_idle >> p->Scell_log) & RED_STAB_MASK];
205 if (shift)
206 return p->qavg >> shift;
207 else {
208 /* Approximate initial part of exponent with linear function:
210 * (1-W)^m ~= 1-mW + ...
212 * Seems, it is the best solution to
213 * problem of too coarse exponent tabulation.
215 us_idle = (p->qavg * (u64)us_idle) >> p->Scell_log;
217 if (us_idle < (p->qavg >> 1))
218 return p->qavg - us_idle;
219 else
220 return p->qavg >> 1;
224 static inline unsigned long red_calc_qavg_no_idle_time(struct red_parms *p,
225 unsigned int backlog)
228 * NOTE: p->qavg is fixed point number with point at Wlog.
229 * The formula below is equvalent to floating point
230 * version:
232 * qavg = qavg*(1-W) + backlog*W;
234 * --ANK (980924)
236 return p->qavg + (backlog - (p->qavg >> p->Wlog));
239 static inline unsigned long red_calc_qavg(struct red_parms *p,
240 unsigned int backlog)
242 if (!red_is_idling(p))
243 return red_calc_qavg_no_idle_time(p, backlog);
244 else
245 return red_calc_qavg_from_idle_time(p);
248 static inline u32 red_random(struct red_parms *p)
250 return net_random() & p->Rmask;
253 static inline int red_mark_probability(struct red_parms *p, unsigned long qavg)
255 /* The formula used below causes questions.
257 OK. qR is random number in the interval 0..Rmask
258 i.e. 0..(2^Plog). If we used floating point
259 arithmetics, it would be: (2^Plog)*rnd_num,
260 where rnd_num is less 1.
262 Taking into account, that qavg have fixed
263 point at Wlog, and Plog is related to max_P by
264 max_P = (qth_max-qth_min)/2^Plog; two lines
265 below have the following floating point equivalent:
267 max_P*(qavg - qth_min)/(qth_max-qth_min) < rnd/qcount
269 Any questions? --ANK (980924)
271 return !(((qavg - p->qth_min) >> p->Wlog) * p->qcount < p->qR);
274 enum {
275 RED_BELOW_MIN_THRESH,
276 RED_BETWEEN_TRESH,
277 RED_ABOVE_MAX_TRESH,
280 static inline int red_cmp_thresh(struct red_parms *p, unsigned long qavg)
282 if (qavg < p->qth_min)
283 return RED_BELOW_MIN_THRESH;
284 else if (qavg >= p->qth_max)
285 return RED_ABOVE_MAX_TRESH;
286 else
287 return RED_BETWEEN_TRESH;
290 enum {
291 RED_DONT_MARK,
292 RED_PROB_MARK,
293 RED_HARD_MARK,
296 static inline int red_action(struct red_parms *p, unsigned long qavg)
298 switch (red_cmp_thresh(p, qavg)) {
299 case RED_BELOW_MIN_THRESH:
300 p->qcount = -1;
301 return RED_DONT_MARK;
303 case RED_BETWEEN_TRESH:
304 if (++p->qcount) {
305 if (red_mark_probability(p, qavg)) {
306 p->qcount = 0;
307 p->qR = red_random(p);
308 return RED_PROB_MARK;
310 } else
311 p->qR = red_random(p);
313 return RED_DONT_MARK;
315 case RED_ABOVE_MAX_TRESH:
316 p->qcount = -1;
317 return RED_HARD_MARK;
320 BUG();
321 return RED_DONT_MARK;
324 #endif