/* Random Early Detection (RED) algorithm. =======================================
Source: Sally Floyd and Van Jacobson, "Random Early Detection Gateways for Congestion Avoidance", 1993, IEEE/ACM Transactions on Networking.
This file codes a "divisionless" version of RED algorithm as written down in Fig.17 of the paper.
Short description. ------------------
When a new packet arrives we calculate the average queue length:
avg = (1-W)*avg + W*current_queue_len,
W is the filter time constant (chosen as 2^(-Wlog)), it controls the inertia of the algorithm. To allow larger bursts, W should be decreased.
if (avg > th_max) -> packet marked (dropped). if (avg < th_min) -> packet passes. if (th_min < avg < th_max) we calculate probability:
Pb = max_P * (avg - th_min)/(th_max-th_min)
and mark (drop) packet with this probability. Pb changes from 0 (at avg==th_min) to max_P (avg==th_max). max_P should be small (not 1), usually 0.01..0.02 is good value.
max_P is chosen as a number, so that max_P/(th_max-th_min) is a negative power of two in order arithmetic to contain only shifts.
Parameters, settable by user: -----------------------------
qth_min - bytes (should be < qth_max/2) qth_max - bytes (should be at least 2*qth_min and less limit) Wlog - bits (<32) log(1/W). Plog - bits (<32)
Plog is related to max_P by formula:
max_P = (qth_max-qth_min)/2^Plog;
F.e. if qth_max=128K and qth_min=32K, then Plog=22 corresponds to max_P=0.02
Scell_log Stab
Lookup table for log((1-W)^(t/t_ave).
NOTES:
Upper bound on W. -----------------
If you want to allow bursts of L packets of size S, you should choose W:
/* * Adaptative RED : An Algorithm for Increasing the Robustness of RED's AQM * (Sally FLoyd, Ramakrishna Gummadi, and Scott Shenker) August 2001 * * Every 500 ms: * if (avg > target and max_p <= 0.5) * increase max_p : max_p += alpha; * else if (avg < target and max_p >= 0.01) * decrease max_p : max_p *= beta; * * target :[qth_min + 0.4*(qth_min - qth_max), * qth_min + 0.6*(qth_min - qth_max)]. * alpha : min(0.01, max_p / 4) * beta : 0.9 * max_P is a Q0.32 fixed point number (with 32 bits mantissa) * max_P between 0.01 and 0.5 (1% - 50%) [ Its no longer a negative power of two ]
*/ #define RED_ONE_PERCENT ((u32)DIV_ROUND_CLOSEST(1ULL<<32, 100))
staticinlinevoid red_set_vars(struct red_vars *v)
{ /* Reset average queue length, the value is strictly bound * to the parameters below, resetting hurts a bit but leaving * it might result in an unreasonable qavg for a while. --TGR
*/
v->qavg = 0;
v->qcount = -1;
}
staticinlinebool red_check_params(u32 qth_min, u32 qth_max, u8 Wlog,
u8 Scell_log, u8 *stab)
{ if (fls(qth_min) + Wlog >= 32) returnfalse; if (fls(qth_max) + Wlog >= 32) returnfalse; if (Scell_log >= 32) returnfalse; if (qth_max < qth_min) returnfalse; if (stab) { int i;
for (i = 0; i < RED_STAB_SIZE; i++) if (stab[i] >= 32) returnfalse;
} returntrue;
}
if (qopt_flags && flags_attr) {
NL_SET_ERR_MSG_MOD(extack, "flags should be passed either through qopt, or through a dedicated attribute"); return -EINVAL;
}
/* * The problem: ideally, average length queue recalculation should * be done over constant clock intervals. This is too expensive, so * that the calculation is driven by outgoing packets. * When the queue is idle we have to model this clock by hand. * * SF+VJ proposed to "generate": * * m = idletime / (average_pkt_size / bandwidth) * * dummy packets as a burst after idle time, i.e. * * v->qavg *= (1-W)^m * * This is an apparently overcomplicated solution (f.e. we have to * precompute a table to make this calculation in reasonable time) * I believe that a simpler model may be used here, * but it is field for experiments.
*/
if (shift) return v->qavg >> shift; else { /* Approximate initial part of exponent with linear function: * * (1-W)^m ~= 1-mW + ... * * Seems, it is the best solution to * problem of too coarse exponent tabulation.
*/
us_idle = (v->qavg * (u64)us_idle) >> p->Scell_log;
staticinlineunsignedlong red_calc_qavg_no_idle_time(conststruct red_parms *p, conststruct red_vars *v, unsignedint backlog)
{ /* * NOTE: v->qavg is fixed point number with point at Wlog. * The formula below is equivalent to floating point * version: * * qavg = qavg*(1-W) + backlog*W; * * --ANK (980924)
*/ return v->qavg + (backlog - (v->qavg >> p->Wlog));
}
staticinlineint red_mark_probability(conststruct red_parms *p, conststruct red_vars *v, unsignedlong qavg)
{ /* The formula used below causes questions.
OK. qR is random number in the interval (0..1/max_P)*(qth_max-qth_min) i.e. 0..(2^Plog). If we used floating point arithmetic, it would be: (2^Plog)*rnd_num, where rnd_num is less 1.
Taking into account, that qavg have fixed point at Wlog, two lines below have the following floating point equivalent:
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.