2 This file is part of GNUnet
3 Copyright (C) 2013 GNUnet e.V.
5 GNUnet is free software: you can redistribute it and/or modify it
6 under the terms of the GNU Affero General Public License as published
7 by the Free Software Foundation, either version 3 of the License,
8 or (at your option) any later version.
10 GNUnet is distributed in the hope that it will be useful, but
11 WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 Affero General Public License for more details.
15 You should have received a copy of the GNU Affero General Public License
16 along with this program. If not, see <http://www.gnu.org/licenses/>.
21 * @brief implementation of simulation for invertible bloom filter
22 * @author Florian Dold
24 * This code was used for some internal experiments, it is not
25 * build or shipped as part of the GNUnet system.
31 #define MAX_IBF_DECODE 16
33 /* report average over how many rounds? */
36 /* enable one of the three below */
39 // possibly slightly better fix for large IBF_DECODE values
47 // avoid assembly? (ASM is about 50% faster)
51 main(int argc, char **argv)
54 unsigned int buckets[31]; // max is 2^31 as 'random' returns only between 0 and 2^31
59 unsigned long long total;
63 srandom (time (NULL));
65 want = atoi (argv[1]);
66 for (round=0;round<ROUNDS;round++)
68 memset (buckets, 0, sizeof (buckets));
71 /* FIXME: might want to use 'better' PRNG to avoid
72 PRNG-induced biases */
77 for (j=0;(j < 31) && (0 == (r & (1 << j)));j++) ;
79 /* use assembly / gcc */
80 j = __builtin_ffs (r) - 1;
86 for (j=31;j >= 0; j--)
89 /* improved algorithm, for 1000 elements with IBF-DECODE 8, I
90 get 990/1000 elements on average over 1 million runs; key
91 idea being to stop short of the 'last' possible IBF as
92 otherwise a "lowball" per-chance would unduely influence the
95 (buckets[j - 1] > MAX_IBF_DECODE) )
97 ret *= (1 << (j + 1));
102 /* another improvement: don't just always cut off the last one,
103 but rather try to predict based on all previous values where
104 that "last" one is; additional prediction can only really
105 work if MAX_IBF_DECODE is sufficiently high */
107 ( (buckets[j - 1] > MAX_IBF_DECODE) ||
108 (predict > MAX_IBF_DECODE) ) )
110 ret *= (1 << (j + 1));
115 /* original algorithm, for 1000 elements with IBF-DECODE 8,
116 I get 920/1000 elements on average over 1 million runs */
117 if (buckets[j] > MAX_IBF_DECODE)
124 predict = (buckets[j] + 2.0 * predict) / 2.0;
127 fprintf (stderr, "%u ", ret);
131 fprintf (stderr, "\n");
132 fprintf (stdout, "average %llu\n", total / ROUNDS);
136 /* TODO: should calculate stddev of the results to also be able to
137 say something about the stability of the results, outside of
138 large-scale averages -- gaining 8% precision at the expense of
139 50% additional variance might not be worth it... */