X-Git-Url: http://git.ozlabs.org/?a=blobdiff_plain;ds=inline;f=ccan%2Fhtable%2Ftools%2Fstringspeed.c;fp=ccan%2Fhtable%2Ftools%2Fstringspeed.c;h=00ad2790cd4cc13e7c58d9ea86f4152d47f7d89d;hb=5c559e7df1d31b4c0ddf26451fac972dc8a0c2c9;hp=0000000000000000000000000000000000000000;hpb=ab83de953730f5e5e571dbf69ffb3cc685a102dc;p=ccan

diff --git a/ccan/htable/tools/stringspeed.c b/ccan/htable/tools/stringspeed.c
new file mode 100644
index 00000000..00ad2790
--- /dev/null
+++ b/ccan/htable/tools/stringspeed.c
@@ -0,0 +1,246 @@
+/* Simple speed tests for a hash of strings. */
+#include <ccan/htable/htable_type.h>
+#include <ccan/htable/htable.c>
+#include <ccan/str_talloc/str_talloc.h>
+#include <ccan/grab_file/grab_file.h>
+#include <ccan/talloc/talloc.h>
+#include <ccan/hash/hash.h>
+#include <ccan/time/time.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <time.h>
+#include <unistd.h>
+#include <sys/time.h>
+
+static size_t hashcount;
+
+static const char *strkey(const char *str)
+{
+	return str;
+}
+
+static size_t hash_str(const char *key)
+{
+	hashcount++;
+	return hash(key, strlen(key), 0);
+}
+
+static bool cmp(const char *obj, const char *key)
+{
+	return strcmp(obj, key) == 0;
+}
+
+HTABLE_DEFINE_TYPE(char, strkey, hash_str, cmp, str);
+
+/* Nanoseconds per operation */
+static size_t normalize(const struct timeval *start,
+			const struct timeval *stop,
+			unsigned int num)
+{
+	struct timeval diff;
+
+	timersub(stop, start, &diff);
+
+	/* Floating point is more accurate here. */
+	return (double)(diff.tv_sec * 1000000 + diff.tv_usec)
+		/ num * 1000;
+}
+
+int main(int argc, char *argv[])
+{
+	size_t i, j, num;
+	struct timeval start, stop;
+	struct htable_str *ht;
+	char **words, **misswords;
+
+	words = strsplit(NULL, grab_file(NULL,
+					 argv[1] ? argv[1] : "/usr/share/dict/words",
+					 NULL), "\n");
+	ht = htable_str_new();
+	num = talloc_array_length(words) - 1;
+	printf("%zu words\n", num);
+
+	/* Append and prepend last char for miss testing. */
+	misswords = talloc_array(words, char *, num);
+	for (i = 0; i < num; i++) {
+		char lastc;
+		if (strlen(words[i]))
+			lastc = words[i][strlen(words[i])-1];
+		else
+			lastc = 'z';
+		misswords[i] = talloc_asprintf(misswords, "%c%s%c%c",
+					       lastc, words[i], lastc, lastc);
+	}
+
+	printf("#01: Initial insert: ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++)
+		htable_str_add(ht, words[i]);
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("Bytes allocated: %zu\n",
+	       sizeof(((struct htable *)ht)->table[0])
+	       << ((struct htable *)ht)->bits);
+
+	printf("#02: Initial lookup (match): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++)
+		if (htable_str_get(ht, words[i]) != words[i])
+			abort();
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("#03: Initial lookup (miss): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++) {
+		if (htable_str_get(ht, misswords[i]))
+			abort();
+	}
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	/* Lookups in order are very cache-friendly for judy; try random */
+	printf("#04: Initial lookup (random): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0, j = 0; i < num; i++, j = (j + 10007) % num)
+		if (htable_str_get(ht, words[j]) != words[j])
+			abort();
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	hashcount = 0;
+	printf("#05: Initial delete all: ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++)
+		if (!htable_str_del(ht, words[i]))
+			abort();
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("#06: Initial re-inserting: ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++)
+		htable_str_add(ht, words[i]);
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	hashcount = 0;
+	printf("#07: Deleting first half: ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i+=2)
+		if (!htable_str_del(ht, words[i]))
+			abort();
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("#08: Adding (a different) half: ");
+	fflush(stdout);
+
+	start = time_now();
+	for (i = 0; i < num; i+=2)
+		htable_str_add(ht, misswords[i]);
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("#09: Lookup after half-change (match): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 1; i < num; i+=2)
+		if (htable_str_get(ht, words[i]) != words[i])
+			abort();
+	for (i = 0; i < num; i+=2) {
+		if (htable_str_get(ht, misswords[i]) != misswords[i])
+			abort();
+	}
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("#10: Lookup after half-change (miss): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i+=2)
+		if (htable_str_get(ht, words[i]))
+			abort();
+	for (i = 1; i < num; i+=2) {
+		if (htable_str_get(ht, misswords[i]))
+			abort();
+	}
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	/* Hashtables with delete markers can fill with markers over time.
+	 * so do some changes to see how it operates in long-term. */
+	printf("#11: Churn 1: ");
+	start = time_now();
+	for (j = 0; j < num; j+=2) {
+		if (!htable_str_del(ht, misswords[j]))
+			abort();
+		if (!htable_str_add(ht, words[j]))
+			abort();
+	}
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("#12: Churn 2: ");
+	start = time_now();
+	for (j = 1; j < num; j+=2) {
+		if (!htable_str_del(ht, words[j]))
+			abort();
+		if (!htable_str_add(ht, misswords[j]))
+			abort();
+	}
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("#13: Churn 3: ");
+	start = time_now();
+	for (j = 1; j < num; j+=2) {
+		if (!htable_str_del(ht, misswords[j]))
+			abort();
+		if (!htable_str_add(ht, words[j]))
+			abort();
+	}
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	/* Now it's back to normal... */
+	printf("#14: Post-Churn lookup (match): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++)
+		if (htable_str_get(ht, words[i]) != words[i])
+			abort();
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("#15: Post-Churn lookup (miss): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++) {
+		if (htable_str_get(ht, misswords[i]))
+			abort();
+	}
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	/* Lookups in order are very cache-friendly for judy; try random */
+	printf("#16: Post-Churn lookup (random): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0, j = 0; i < num; i++, j = (j + 10007) % num)
+		if (htable_str_get(ht, words[j]) != words[j])
+			abort();
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	return 0;
+}