progress

4 files changed, 745 insertions, 0 deletions

diff --git a/ccan/htable/tools/Makefile b/ccan/htable/tools/Makefile
new file mode 100644
index 0000000..a2cad59
--- /dev/null
+++ b/ccan/htable/tools/Makefile
@@ -0,0 +1,40 @@
+CCANDIR=../../..
+CFLAGS=-Wall -Werror -O3 -I$(CCANDIR)
+#CFLAGS=-Wall -Werror -g -I$(CCANDIR)
+
+CCAN_OBJS:=ccan-tal.o ccan-tal-str.o ccan-tal-grab_file.o ccan-take.o ccan-time.o ccan-str.o ccan-noerr.o ccan-list.o
+
+all: speed stringspeed hsearchspeed
+
+speed: speed.o hash.o $(CCAN_OBJS)
+
+speed.o: speed.c ../htable.h ../htable.c
+
+hash.o: ../../hash/hash.c
+	$(CC) $(CFLAGS) -c -o $@ $<
+
+stringspeed: stringspeed.o hash.o $(CCAN_OBJS)
+
+stringspeed.o: speed.c ../htable.h ../htable.c
+
+hsearchspeed: hsearchspeed.o $(CCAN_OBJS)
+
+clean:
+	rm -f stringspeed speed hsearchspeed *.o
+
+ccan-tal.o: $(CCANDIR)/ccan/tal/tal.c
+	$(CC) $(CFLAGS) -c -o $@ $<
+ccan-tal-str.o: $(CCANDIR)/ccan/tal/str/str.c
+	$(CC) $(CFLAGS) -c -o $@ $<
+ccan-take.o: $(CCANDIR)/ccan/take/take.c
+	$(CC) $(CFLAGS) -c -o $@ $<
+ccan-tal-grab_file.o: $(CCANDIR)/ccan/tal/grab_file/grab_file.c
+	$(CC) $(CFLAGS) -c -o $@ $<
+ccan-time.o: $(CCANDIR)/ccan/time/time.c
+	$(CC) $(CFLAGS) -c -o $@ $<
+ccan-list.o: $(CCANDIR)/ccan/list/list.c
+	$(CC) $(CFLAGS) -c -o $@ $<
+ccan-str.o: $(CCANDIR)/ccan/str/str.c
+	$(CC) $(CFLAGS) -c -o $@ $<
+ccan-noerr.o: $(CCANDIR)/ccan/noerr/noerr.c
+	$(CC) $(CFLAGS) -c -o $@ $<
diff --git a/ccan/htable/tools/hsearchspeed.c b/ccan/htable/tools/hsearchspeed.c
new file mode 100644
index 0000000..8828011
--- /dev/null
+++ b/ccan/htable/tools/hsearchspeed.c
@@ -0,0 +1,95 @@
+/* Simple speed tests for a hash of strings using hsearch */
+#include <ccan/htable/htable_type.h>
+#include <ccan/htable/htable.c>
+#include <ccan/tal/str/str.h>
+#include <ccan/tal/grab_file/grab_file.h>
+#include <ccan/tal/tal.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>
+#include <search.h>
+
+/* Nanoseconds per operation */
+static size_t normalize(const struct timeabs *start,
+			const struct timeabs *stop,
+			unsigned int num)
+{
+	return time_to_nsec(time_divide(time_between(*stop, *start), num));
+}
+
+int main(int argc, char *argv[])
+{
+	size_t i, j, num;
+	struct timeabs start, stop;
+	char **w;
+	ENTRY *words, *misswords;
+
+	w = tal_strsplit(NULL, grab_file(NULL,
+					 argv[1] ? argv[1] : "/usr/share/dict/words"), "\n", STR_NO_EMPTY);
+	num = tal_count(w) - 1;
+	printf("%zu words\n", num);
+
+	hcreate(num+num/3);
+
+	words = tal_arr(w, ENTRY, num);
+	for (i = 0; i < num; i++) {
+		words[i].key = w[i];
+		words[i].data = words[i].key;
+	}
+
+	/* Append and prepend last char for miss testing. */
+	misswords = tal_arr(w, ENTRY, num);
+	for (i = 0; i < num; i++) {
+		char lastc;
+		if (strlen(w[i]))
+			lastc = w[i][strlen(w[i])-1];
+		else
+			lastc = 'z';
+		misswords[i].key = tal_fmt(misswords, "%c%s%c%c",
+					   lastc, w[i], lastc, lastc);
+	}
+
+	printf("#01: Initial insert: ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++)
+		hsearch(words[i], ENTER);
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("#02: Initial lookup (match): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++)
+		if (hsearch(words[i], FIND)->data != words[i].data)
+			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 (hsearch(misswords[i], FIND))
+			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 (hsearch(words[i], FIND)->data != words[i].data)
+			abort();
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	return 0;
+}
diff --git a/ccan/htable/tools/speed.c b/ccan/htable/tools/speed.c
new file mode 100644
index 0000000..dce3fdf
--- /dev/null
+++ b/ccan/htable/tools/speed.c
@@ -0,0 +1,370 @@
+/* Simple speed tests for hashtables. */
+#include <ccan/htable/htable_type.h>
+#include <ccan/htable/htable.c>
+#include <ccan/hash/hash.h>
+#include <ccan/time/time.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+static size_t hashcount;
+struct object {
+	/* The key. */
+	unsigned int key;
+
+	/* Some contents. Doubles as consistency check. */
+	struct object *self;
+};
+
+static const unsigned int *objkey(const struct object *obj)
+{
+	return &obj->key;
+}
+
+static size_t hash_obj(const unsigned int *key)
+{
+	hashcount++;
+	return hashl(key, 1, 0);
+}
+
+static bool cmp(const struct object *object, const unsigned int *key)
+{
+	return object->key == *key;
+}
+
+HTABLE_DEFINE_TYPE(struct object, objkey, hash_obj, cmp, htable_obj);
+
+static unsigned int popcount(unsigned long val)
+{
+#if HAVE_BUILTIN_POPCOUNTL
+	return __builtin_popcountl(val);
+#else
+	if (sizeof(long) == sizeof(u64)) {
+		u64 v = val;
+		v = (v & 0x5555555555555555ULL)
+			+ ((v >> 1) & 0x5555555555555555ULL);
+		v = (v & 0x3333333333333333ULL)
+			+ ((v >> 1) & 0x3333333333333333ULL);
+		v = (v & 0x0F0F0F0F0F0F0F0FULL)
+			+ ((v >> 1) & 0x0F0F0F0F0F0F0F0FULL);
+		v = (v & 0x00FF00FF00FF00FFULL)
+			+ ((v >> 1) & 0x00FF00FF00FF00FFULL);
+		v = (v & 0x0000FFFF0000FFFFULL)
+			+ ((v >> 1) & 0x0000FFFF0000FFFFULL);
+		v = (v & 0x00000000FFFFFFFFULL)
+			+ ((v >> 1) & 0x00000000FFFFFFFFULL);
+		return v;
+	}
+	val = (val & 0x55555555ULL) + ((val >> 1) & 0x55555555ULL);
+	val = (val & 0x33333333ULL) + ((val >> 1) & 0x33333333ULL);
+	val = (val & 0x0F0F0F0FULL) + ((val >> 1) & 0x0F0F0F0FULL);
+	val = (val & 0x00FF00FFULL) + ((val >> 1) & 0x00FF00FFULL);
+	val = (val & 0x0000FFFFULL) + ((val >> 1) & 0x0000FFFFULL);
+	return val;
+#endif
+}
+
+static size_t perfect(const struct htable *ht)
+{
+	size_t i, placed_perfect = 0;
+
+	for (i = 0; i < ((size_t)1 << ht->bits); i++) {
+		if (!entry_is_valid(ht->table[i]))
+			continue;
+		if (hash_bucket(ht, ht->rehash(get_raw_ptr(ht, ht->table[i]),
+					       ht->priv)) == i) {
+			assert((ht->table[i] & ht->perfect_bit)
+			       == ht->perfect_bit);
+			placed_perfect++;
+		}
+	}
+	return placed_perfect;
+}
+
+static size_t count_deleted(const struct htable *ht)
+{
+	size_t i, delete_markers = 0;
+
+	for (i = 0; i < ((size_t)1 << ht->bits); i++) {
+		if (ht->table[i] == HTABLE_DELETED)
+			delete_markers++;
+	}
+	return delete_markers;
+}
+
+/* Nanoseconds per operation */
+static size_t normalize(const struct timeabs *start,
+			const struct timeabs *stop,
+			unsigned int num)
+{
+	return time_to_nsec(time_divide(time_between(*stop, *start), num));
+}
+
+static size_t worst_run(struct htable *ht, size_t *deleted)
+{
+	size_t longest = 0, len = 0, this_del = 0, i;
+
+	*deleted = 0;
+	/* This doesn't take into account end-wrap, but gives an idea. */
+	for (i = 0; i < ((size_t)1 << ht->bits); i++) {
+		if (ht->table[i]) {
+			len++;
+			if (ht->table[i] == HTABLE_DELETED)
+				this_del++;
+		} else {
+			if (len > longest) {
+				longest = len;
+				*deleted = this_del;
+			}
+			len = 0;
+			this_del = 0;
+		}
+	}
+	return longest;
+}
+
+int main(int argc, char *argv[])
+{
+	struct object *objs;
+	unsigned int i, j;
+	size_t num, deleted;
+	struct timeabs start, stop;
+	struct htable_obj ht;
+	bool make_dumb = false;
+
+	if (argv[1] && strcmp(argv[1], "--dumb") == 0) {
+		argv++;
+		make_dumb = true;
+	}
+	num = argv[1] ? atoi(argv[1]) : 1000000;
+	objs = calloc(num, sizeof(objs[0]));
+
+	for (i = 0; i < num; i++) {
+		objs[i].key = i;
+		objs[i].self = &objs[i];
+	}
+
+	htable_obj_init(&ht);
+
+	printf("Initial insert: ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++)
+		htable_obj_add(&ht, objs[i].self);
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+	printf("Details: hash size %u, mask bits %u, perfect %.0f%%\n",
+	       1U << ht.raw.bits, popcount(ht.raw.common_mask),
+	       perfect(&ht.raw) * 100.0 / ht.raw.elems);
+
+	if (make_dumb) {
+		/* Screw with mask, to hobble us. */
+		update_common(&ht.raw, (void *)~ht.raw.common_bits);
+		printf("Details: DUMB MODE: mask bits %u\n",
+		       popcount(ht.raw.common_mask));
+	}
+
+	printf("Initial lookup (match): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++)
+		if (htable_obj_get(&ht, &i)->self != objs[i].self)
+			abort();
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("Initial lookup (miss): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++) {
+		unsigned int n = i + num;
+		if (htable_obj_get(&ht, &n))
+			abort();
+	}
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	/* Lookups in order are very cache-friendly for judy; try random */
+	printf("Initial lookup (random): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0, j = 0; i < num; i++, j = (j + 10007) % num)
+		if (htable_obj_get(&ht, &j)->self != &objs[j])
+			abort();
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	hashcount = 0;
+	printf("Initial delete all: ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++)
+		if (!htable_obj_del(&ht, objs[i].self))
+			abort();
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+	printf("Details: rehashes %zu\n", hashcount);
+
+	printf("Initial re-inserting: ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++)
+		htable_obj_add(&ht, objs[i].self);
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	hashcount = 0;
+	printf("Deleting first half: ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i+=2)
+		if (!htable_obj_del(&ht, objs[i].self))
+			abort();
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("Details: rehashes %zu, delete markers %zu\n",
+	       hashcount, count_deleted(&ht.raw));
+
+	printf("Adding (a different) half: ");
+	fflush(stdout);
+
+	for (i = 0; i < num; i+=2)
+		objs[i].key = num+i;
+
+	start = time_now();
+	for (i = 0; i < num; i+=2)
+		htable_obj_add(&ht, objs[i].self);
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("Details: delete markers %zu, perfect %.0f%%\n",
+	       count_deleted(&ht.raw), perfect(&ht.raw) * 100.0 / ht.raw.elems);
+
+	printf("Lookup after half-change (match): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 1; i < num; i+=2)
+		if (htable_obj_get(&ht, &i)->self != objs[i].self)
+			abort();
+	for (i = 0; i < num; i+=2) {
+		unsigned int n = i + num;
+		if (htable_obj_get(&ht, &n)->self != objs[i].self)
+			abort();
+	}
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("Lookup after half-change (miss): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++) {
+		unsigned int n = i + num * 2;
+		if (htable_obj_get(&ht, &n))
+			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. */
+	for (i = 0; i < 5; i++) {
+		if (i == 0) {
+			/* We don't measure this: jmap is different. */
+			printf("Details: initial churn\n");
+		} else {
+			printf("Churning %s time: ",
+			       i == 1 ? "second"
+			       : i == 2 ? "third"
+			       : i == 3 ? "fourth"
+			       : "fifth");
+			fflush(stdout);
+		}
+		start = time_now();
+		for (j = 0; j < num; j++) {
+			if (!htable_obj_del(&ht, &objs[j]))
+				abort();
+			objs[j].key = num*i+j;
+			if (!htable_obj_add(&ht, &objs[j]))
+				abort();
+		}
+		stop = time_now();
+		if (i != 0)
+			printf(" %zu ns\n", normalize(&start, &stop, num));
+	}
+
+	/* Spread out the keys more to try to make it harder. */
+	printf("Details: reinserting with spread\n");
+	for (i = 0; i < num; i++) {
+		if (!htable_obj_del(&ht, objs[i].self))
+			abort();
+		objs[i].key = num * 5 + i * 9;
+		if (!htable_obj_add(&ht, objs[i].self))
+			abort();
+	}
+	printf("Details: delete markers %zu, perfect %.0f%%\n",
+	       count_deleted(&ht.raw), perfect(&ht.raw) * 100.0 / ht.raw.elems);
+	i = worst_run(&ht.raw, &deleted);
+	printf("Details: worst run %u (%zu deleted)\n", i, deleted);
+
+	printf("Lookup after churn & spread (match): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++) {
+		unsigned int n = num * 5 + i * 9;
+		if (htable_obj_get(&ht, &n)->self != objs[i].self)
+			abort();
+	}
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("Lookup after churn & spread (miss): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i++) {
+		unsigned int n = num * (5 + 9) + i * 9;
+		if (htable_obj_get(&ht, &n))
+			abort();
+	}
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("Lookup after churn & spread (random): ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0, j = 0; i < num; i++, j = (j + 10007) % num) {
+		unsigned int n = num * 5 + j * 9;
+		if (htable_obj_get(&ht, &n)->self != &objs[j])
+			abort();
+	}
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	hashcount = 0;
+	printf("Deleting half after churn & spread: ");
+	fflush(stdout);
+	start = time_now();
+	for (i = 0; i < num; i+=2)
+		if (!htable_obj_del(&ht, objs[i].self))
+			abort();
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("Adding (a different) half after churn & spread: ");
+	fflush(stdout);
+
+	for (i = 0; i < num; i+=2)
+		objs[i].key = num*6+i*9;
+
+	start = time_now();
+	for (i = 0; i < num; i+=2)
+		htable_obj_add(&ht, objs[i].self);
+	stop = time_now();
+	printf(" %zu ns\n", normalize(&start, &stop, num));
+
+	printf("Details: delete markers %zu, perfect %.0f%%\n",
+	       count_deleted(&ht.raw), perfect(&ht.raw) * 100.0 / ht.raw.elems);
+
+	return 0;
+}
diff --git a/ccan/htable/tools/stringspeed.c b/ccan/htable/tools/stringspeed.c
new file mode 100644
index 0000000..c6ca10f
--- /dev/null
+++ b/ccan/htable/tools/stringspeed.c
@@ -0,0 +1,240 @@
+/* Simple speed tests for a hash of strings. */
+#include <ccan/htable/htable_type.h>
+#include <ccan/htable/htable.c>
+#include <ccan/tal/str/str.h>
+#include <ccan/tal/grab_file/grab_file.h>
+#include <ccan/tal/tal.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, htable_str);
+
+/* Nanoseconds per operation */
+static size_t normalize(const struct timeabs *start,
+			const struct timeabs *stop,
+			unsigned int num)
+{
+	return time_to_nsec(time_divide(time_between(*stop, *start), num));
+}
+
+int main(int argc, char *argv[])
+{
+	size_t i, j, num;
+	struct timeabs start, stop;
+	struct htable_str ht;
+	char **words, **misswords;
+
+	words = tal_strsplit(NULL, grab_file(NULL,
+					     argv[1] ? argv[1] : "/usr/share/dict/words"), "\n",
+			     STR_NO_EMPTY);
+	htable_str_init(&ht);
+	num = tal_count(words) - 1;
+	/* Note that on my system, num is just > 98304, where we double! */
+	printf("%zu words\n", num);
+
+	/* Append and prepend last char for miss testing. */
+	misswords = tal_arr(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] = tal_fmt(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(ht.raw.table[0]) << ht.raw.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;
+}