dedupe: default to using a bloom filter to save memory
With the bloom filter, we can comfortably check devices up to
the petabyte range, which isn't feasible if we track each
extent.
Signed-off-by: Jens Axboe <axboe@fb.com>
diff --git a/t/dedupe.c b/t/dedupe.c
index b81e98a..ac5d7f2 100644
--- a/t/dedupe.c
+++ b/t/dedupe.c
@@ -27,6 +27,8 @@
#include "../gettime.h"
#include "../fio_time.h"
+#include "../lib/bloom.h"
+
FILE *f_err;
struct timeval *fio_tv = NULL;
unsigned int fio_debug = 0;
@@ -45,6 +47,7 @@
uint64_t size;
unsigned long items;
+ unsigned long dupes;
int err;
};
@@ -66,6 +69,7 @@
};
static struct rb_root rb_root;
+static struct bloom *bloom;
static struct fio_mutex *rb_lock;
static unsigned int blocksize = 4096;
@@ -75,6 +79,7 @@
static unsigned int odirect;
static unsigned int collision_check;
static unsigned int print_progress = 1;
+static unsigned int use_bloom = 1;
static uint64_t total_size;
static uint64_t cur_offset;
@@ -231,14 +236,24 @@
add_item(c, i);
}
-static void insert_chunks(struct item *items, unsigned int nitems)
+static void insert_chunks(struct item *items, unsigned int nitems,
+ uint64_t *ndupes)
{
int i;
fio_mutex_down(rb_lock);
- for (i = 0; i < nitems; i++)
- insert_chunk(&items[i]);
+ for (i = 0; i < nitems; i++) {
+ if (bloom) {
+ unsigned int s;
+ int r;
+
+ s = sizeof(items[i].hash) / sizeof(uint32_t);
+ r = bloom_set(bloom, items[i].hash, s);
+ *ndupes += r;
+ } else
+ insert_chunk(&items[i]);
+ }
fio_mutex_up(rb_lock);
}
@@ -257,6 +272,7 @@
unsigned int nblocks, i;
off_t offset;
int err = 0, nitems = 0;
+ uint64_t ndupes = 0;
struct item *items;
nblocks = thread->size / blocksize;
@@ -266,15 +282,18 @@
for (i = 0; i < nblocks; i++) {
if (read_block(thread->fd, buf, offset))
break;
- items[i].offset = offset;
+ if (items)
+ items[i].offset = offset;
crc_buf(buf, items[i].hash);
offset += blocksize;
nitems++;
}
- insert_chunks(items, nitems);
- thread->items += nitems;
+ insert_chunks(items, nitems, &ndupes);
+
free(items);
+ thread->items += nitems;
+ thread->dupes += ndupes;
return err;
}
@@ -343,7 +362,8 @@
};
}
-static int run_dedupe_threads(int fd, uint64_t dev_size)
+static int run_dedupe_threads(int fd, uint64_t dev_size, uint64_t *nextents,
+ uint64_t *nchunks)
{
struct worker_thread *threads;
unsigned long nitems, total_items;
@@ -371,20 +391,27 @@
show_progress(threads, total_items);
nitems = 0;
+ *nextents = 0;
+ *nchunks = 1;
for (i = 0; i < num_threads; i++) {
void *ret;
pthread_join(threads[i].thread, &ret);
nitems += threads[i].items;
+ *nchunks += threads[i].dupes;
}
printf("Threads(%u): %lu items processed\n", num_threads, nitems);
+ *nextents = nitems;
+ *nchunks = nitems - *nchunks;
+
fio_mutex_remove(size_lock);
free(threads);
return err;
}
-static int dedupe_check(const char *filename)
+static int dedupe_check(const char *filename, uint64_t *nextents,
+ uint64_t *nchunks)
{
uint64_t dev_size;
struct stat sb;
@@ -412,9 +439,16 @@
return 1;
}
+ if (use_bloom) {
+ uint64_t bloom_entries;
+
+ bloom_entries = (3 * dev_size ) / (blocksize * 2);
+ bloom = bloom_new(bloom_entries);
+ }
+
printf("Will check <%s>, size <%llu>, using %u threads\n", filename, (unsigned long long) dev_size, num_threads);
- return run_dedupe_threads(dev_fd, dev_size);
+ return run_dedupe_threads(dev_fd, dev_size, nextents, nchunks);
}
static void show_chunk(struct chunk *c)
@@ -429,14 +463,24 @@
}
}
-static void iter_rb_tree(void)
+static void show_stat(uint64_t nextents, uint64_t nchunks)
{
- struct rb_node *n;
- uint64_t nchunks;
- uint64_t nextents;
double perc;
- nchunks = nextents = 0;
+ printf("Extents=%lu, Unique extents=%lu\n", (unsigned long) nextents, (unsigned long) nchunks);
+ printf("De-dupe factor: %3.2f\n", (double) nextents / (double) nchunks);
+
+ perc = 1.00 - ((double) nchunks / (double) nextents);
+ perc *= 100.0;
+ printf("Fio setting: dedupe_percentage=%u\n", (int) (perc + 0.50));
+
+}
+
+static void iter_rb_tree(uint64_t *nextents, uint64_t *nchunks)
+{
+ struct rb_node *n;
+
+ *nchunks = *nextents = 0;
n = rb_first(&rb_root);
if (!n)
@@ -446,20 +490,13 @@
struct chunk *c;
c = rb_entry(n, struct chunk, rb_node);
- nchunks++;
- nextents += c->count;
+ (*nchunks)++;
+ *nextents += c->count;
if (dump_output)
show_chunk(c);
} while ((n = rb_next(n)) != NULL);
-
- printf("Extents=%lu, Unique extents=%lu\n", (unsigned long) nextents, (unsigned long) nchunks);
- printf("De-dupe factor: %3.2f\n", (double) nextents / (double) nchunks);
-
- perc = 1.00 - ((double) nchunks / (double) nextents);
- perc *= 100.0;
- printf("Fio setting: dedupe_percentage=%u\n", (int) (perc + 0.50));
}
static int usage(char *argv[])
@@ -471,15 +508,17 @@
log_err("\t-d\tFull extent/chunk debug output\n");
log_err("\t-o\tUse O_DIRECT\n");
log_err("\t-c\tFull collision check\n");
+ log_err("\t-B\tUse probabilistic bloom filter\n");
log_err("\t-p\tPrint progress indicator\n");
return 1;
}
int main(int argc, char *argv[])
{
+ uint64_t nextents, nchunks;
int c, ret;
- while ((c = getopt(argc, argv, "b:t:d:o:c:p:")) != -1) {
+ while ((c = getopt(argc, argv, "b:t:d:o:c:p:B:")) != -1) {
switch (c) {
case 'b':
blocksize = atoi(optarg);
@@ -499,12 +538,18 @@
case 'p':
print_progress = atoi(optarg);
break;
+ case 'B':
+ use_bloom = atoi(optarg);
+ break;
case '?':
default:
return usage(argv);
}
}
+ if (collision_check || dump_output)
+ use_bloom = 0;
+
if (!num_threads)
num_threads = cpus_online();
@@ -516,11 +561,15 @@
rb_root = RB_ROOT;
rb_lock = fio_mutex_init(FIO_MUTEX_UNLOCKED);
- ret = dedupe_check(argv[optind]);
+ ret = dedupe_check(argv[optind], &nextents, &nchunks);
- iter_rb_tree();
+ if (!bloom)
+ iter_rb_tree(&nextents, &nchunks);
+
+ show_stat(nextents, nchunks);
fio_mutex_remove(rb_lock);
+ bloom_free(bloom);
scleanup();
return ret;
}
