nostrdb

an unfairly fast embedded nostr database backed by lmdb
git clone git://jb55.com/nostrdb
Log | Files | Refs | Submodules | README | LICENSE

hsearchspeed.c (2410B)


      1 /* Simple speed tests for a hash of strings using hsearch */
      2 #include <ccan/htable/htable_type.h>
      3 #include <ccan/htable/htable.c>
      4 #include <ccan/tal/str/str.h>
      5 #include <ccan/tal/grab_file/grab_file.h>
      6 #include <ccan/tal/tal.h>
      7 #include <ccan/hash/hash.h>
      8 #include <ccan/time/time.h>
      9 #include <stdio.h>
     10 #include <stdlib.h>
     11 #include <string.h>
     12 #include <time.h>
     13 #include <unistd.h>
     14 #include <sys/time.h>
     15 #include <search.h>
     16 
     17 /* Nanoseconds per operation */
     18 static size_t normalize(const struct timeabs *start,
     19 			const struct timeabs *stop,
     20 			unsigned int num)
     21 {
     22 	return time_to_nsec(time_divide(time_between(*stop, *start), num));
     23 }
     24 
     25 int main(int argc, char *argv[])
     26 {
     27 	size_t i, j, num;
     28 	struct timeabs start, stop;
     29 	char **w;
     30 	ENTRY *words, *misswords;
     31 
     32 	w = tal_strsplit(NULL, grab_file(NULL,
     33 					 argv[1] ? argv[1] : "/usr/share/dict/words"), "\n", STR_NO_EMPTY);
     34 	num = tal_count(w) - 1;
     35 	printf("%zu words\n", num);
     36 
     37 	hcreate(num+num/3);
     38 
     39 	words = tal_arr(w, ENTRY, num);
     40 	for (i = 0; i < num; i++) {
     41 		words[i].key = w[i];
     42 		words[i].data = words[i].key;
     43 	}
     44 
     45 	/* Append and prepend last char for miss testing. */
     46 	misswords = tal_arr(w, ENTRY, num);
     47 	for (i = 0; i < num; i++) {
     48 		char lastc;
     49 		if (strlen(w[i]))
     50 			lastc = w[i][strlen(w[i])-1];
     51 		else
     52 			lastc = 'z';
     53 		misswords[i].key = tal_fmt(misswords, "%c%s%c%c",
     54 					   lastc, w[i], lastc, lastc);
     55 	}
     56 
     57 	printf("#01: Initial insert: ");
     58 	fflush(stdout);
     59 	start = time_now();
     60 	for (i = 0; i < num; i++)
     61 		hsearch(words[i], ENTER);
     62 	stop = time_now();
     63 	printf(" %zu ns\n", normalize(&start, &stop, num));
     64 
     65 	printf("#02: Initial lookup (match): ");
     66 	fflush(stdout);
     67 	start = time_now();
     68 	for (i = 0; i < num; i++)
     69 		if (hsearch(words[i], FIND)->data != words[i].data)
     70 			abort();
     71 	stop = time_now();
     72 	printf(" %zu ns\n", normalize(&start, &stop, num));
     73 
     74 	printf("#03: Initial lookup (miss): ");
     75 	fflush(stdout);
     76 	start = time_now();
     77 	for (i = 0; i < num; i++) {
     78 		if (hsearch(misswords[i], FIND))
     79 			abort();
     80 	}
     81 	stop = time_now();
     82 	printf(" %zu ns\n", normalize(&start, &stop, num));
     83 
     84 	/* Lookups in order are very cache-friendly for judy; try random */
     85 	printf("#04: Initial lookup (random): ");
     86 	fflush(stdout);
     87 	start = time_now();
     88 	for (i = 0, j = 0; i < num; i++, j = (j + 10007) % num)
     89 		if (hsearch(words[i], FIND)->data != words[i].data)
     90 			abort();
     91 	stop = time_now();
     92 	printf(" %zu ns\n", normalize(&start, &stop, num));
     93 
     94 	return 0;
     95 }