The following changes since commit 9f51d89c683d70cd8ab23ba09ec6e628a548af5a:
Sync io_uring header with the kernel (2021-11-20 07:31:20 -0700)
are available in the Git repository at:
git://git.kernel.dk/fio.git master
for you to fetch changes up to 1d08bfb018e600cc47f122fb78c02bf74b84dee8:
t/dedupe: style fixups (2021-11-21 06:51:11 -0700)
----------------------------------------------------------------
Bar David (2):
Mixed dedup and compression
fio-dedup: adjusted the binary to support compression
Jens Axboe (3):
Merge branch 'dedupe_and_compression' of https://github.com/bardavid/fio
t/io_uring: fix 32-bit compile warnings
t/dedupe: style fixups
DEDUPE-TODO | 3 --
dedupe.c | 12 ++++-
io_u.c | 29 ++++++-----
t/dedupe.c | 167 +++++++++++++++++++++++++++++++++++++++++++++++------------
t/io_uring.c | 4 +-
5 files changed, 161 insertions(+), 54 deletions(-)
---
Diff of recent changes:
diff --git a/DEDUPE-TODO b/DEDUPE-TODO
index 1f3ee9da..4b0bfd1d 100644
--- a/DEDUPE-TODO
+++ b/DEDUPE-TODO
@@ -1,6 +1,3 @@
-- Mixed buffers of dedupe-able and compressible data.
- Major usecase in performance benchmarking of storage subsystems.
-
- Shifted dedup-able data.
Allow for dedup buffer generation to shift contents by random number
of sectors (fill the gaps with uncompressible data). Some storage
diff --git a/dedupe.c b/dedupe.c
index 043a376c..fd116dfb 100644
--- a/dedupe.c
+++ b/dedupe.c
@@ -2,12 +2,14 @@
int init_dedupe_working_set_seeds(struct thread_data *td)
{
- unsigned long long i;
+ unsigned long long i, j, num_seed_advancements;
struct frand_state dedupe_working_set_state = {0};
if (!td->o.dedupe_percentage || !(td->o.dedupe_mode == DEDUPE_MODE_WORKING_SET))
return 0;
+ num_seed_advancements = td->o.min_bs[DDIR_WRITE] /
+ min_not_zero(td->o.min_bs[DDIR_WRITE], (unsigned long long) td->o.compress_chunk);
/*
* The dedupe working set keeps seeds of unique data (generated by buf_state).
* Dedupe-ed pages will be generated using those seeds.
@@ -21,7 +23,13 @@ int init_dedupe_working_set_seeds(struct thread_data *td)
frand_copy(&dedupe_working_set_state, &td->buf_state);
for (i = 0; i < td->num_unique_pages; i++) {
frand_copy(&td->dedupe_working_set_states[i], &dedupe_working_set_state);
- __get_next_seed(&dedupe_working_set_state);
+ /*
+ * When compression is used the seed is advanced multiple times to
+ * generate the buffer. We want to regenerate the same buffer when
+ * deduping against this page
+ */
+ for (j = 0; j < num_seed_advancements; j++)
+ __get_next_seed(&dedupe_working_set_state);
}
return 0;
diff --git a/io_u.c b/io_u.c
index 586a4bef..3c72d63d 100644
--- a/io_u.c
+++ b/io_u.c
@@ -2230,27 +2230,30 @@ void fill_io_buffer(struct thread_data *td, void *buf, unsigned long long min_wr
if (o->compress_percentage || o->dedupe_percentage) {
unsigned int perc = td->o.compress_percentage;
- struct frand_state *rs;
+ struct frand_state *rs = NULL;
unsigned long long left = max_bs;
unsigned long long this_write;
do {
- rs = get_buf_state(td);
+ /*
+ * Buffers are either entirely dedupe-able or not.
+ * If we choose to dedup, the buffer should undergo
+ * the same manipulation as the original write. Which
+ * means we should retrack the steps we took for compression
+ * as well.
+ */
+ if (!rs)
+ rs = get_buf_state(td);
min_write = min(min_write, left);
- if (perc) {
- this_write = min_not_zero(min_write,
- (unsigned long long) td->o.compress_chunk);
+ this_write = min_not_zero(min_write,
+ (unsigned long long) td->o.compress_chunk);
- fill_random_buf_percentage(rs, buf, perc,
- this_write, this_write,
- o->buffer_pattern,
- o->buffer_pattern_bytes);
- } else {
- fill_random_buf(rs, buf, min_write);
- this_write = min_write;
- }
+ fill_random_buf_percentage(rs, buf, perc,
+ this_write, this_write,
+ o->buffer_pattern,
+ o->buffer_pattern_bytes);
buf += this_write;
left -= this_write;
diff --git a/t/dedupe.c b/t/dedupe.c
index 8b659c76..109ea1af 100644
--- a/t/dedupe.c
+++ b/t/dedupe.c
@@ -24,19 +24,25 @@
#include "../lib/bloom.h"
#include "debug.h"
+#include "zlib.h"
+
+struct zlib_ctrl {
+ z_stream stream;
+ unsigned char *buf_in;
+ unsigned char *buf_out;
+};
struct worker_thread {
+ struct zlib_ctrl zc;
pthread_t thread;
-
- volatile int done;
-
- int fd;
uint64_t cur_offset;
uint64_t size;
-
+ unsigned long long unique_capacity;
unsigned long items;
unsigned long dupes;
int err;
+ int fd;
+ volatile int done;
};
struct extent {
@@ -68,6 +74,7 @@ static unsigned int odirect;
static unsigned int collision_check;
static unsigned int print_progress = 1;
static unsigned int use_bloom = 1;
+static unsigned int compression = 0;
static uint64_t total_size;
static uint64_t cur_offset;
@@ -87,8 +94,9 @@ static uint64_t get_size(struct fio_file *f, struct stat *sb)
return 0;
}
ret = bytes;
- } else
+ } else {
ret = sb->st_size;
+ }
return (ret & ~((uint64_t)blocksize - 1));
}
@@ -120,9 +128,9 @@ static int __read_block(int fd, void *buf, off_t offset, size_t count)
if (ret < 0) {
perror("pread");
return 1;
- } else if (!ret)
+ } else if (!ret) {
return 1;
- else if (ret != count) {
+ } else if (ret != count) {
log_err("dedupe: short read on block\n");
return 1;
}
@@ -135,6 +143,34 @@ static int read_block(int fd, void *buf, off_t offset)
return __read_block(fd, buf, offset, blocksize);
}
+static void account_unique_capacity(uint64_t offset, uint64_t *unique_capacity,
+ struct zlib_ctrl *zc)
+{
+ z_stream *stream = &zc->stream;
+ unsigned int compressed_len;
+ int ret;
+
+ if (read_block(file.fd, zc->buf_in, offset))
+ return;
+
+ stream->next_in = zc->buf_in;
+ stream->avail_in = blocksize;
+ stream->avail_out = deflateBound(stream, blocksize);
+ stream->next_out = zc->buf_out;
+
+ ret = deflate(stream, Z_FINISH);
+ assert(ret != Z_STREAM_ERROR);
+ compressed_len = blocksize - stream->avail_out;
+
+ if (dump_output)
+ printf("offset 0x%lx compressed to %d blocksize %d ratio %.2f \n",
+ (unsigned long) offset, compressed_len, blocksize,
+ (float)compressed_len / (float)blocksize);
+
+ *unique_capacity += compressed_len;
+ deflateReset(stream);
+}
+
static void add_item(struct chunk *c, struct item *i)
{
/*
@@ -182,13 +218,15 @@ static struct chunk *alloc_chunk(void)
if (collision_check || dump_output) {
c = malloc(sizeof(struct chunk) + sizeof(struct flist_head));
INIT_FLIST_HEAD(&c->extent_list[0]);
- } else
+ } else {
c = malloc(sizeof(struct chunk));
+ }
return c;
}
-static void insert_chunk(struct item *i)
+static void insert_chunk(struct item *i, uint64_t *unique_capacity,
+ struct zlib_ctrl *zc)
{
struct fio_rb_node **p, *parent;
struct chunk *c;
@@ -201,11 +239,11 @@ static void insert_chunk(struct item *i)
c = rb_entry(parent, struct chunk, rb_node);
diff = memcmp(i->hash, c->hash, sizeof(i->hash));
- if (diff < 0)
+ if (diff < 0) {
p = &(*p)->rb_left;
- else if (diff > 0)
+ } else if (diff > 0) {
p = &(*p)->rb_right;
- else {
+ } else {
int ret;
if (!collision_check)
@@ -228,12 +266,15 @@ static void insert_chunk(struct item *i)
memcpy(c->hash, i->hash, sizeof(i->hash));
rb_link_node(&c->rb_node, parent, p);
rb_insert_color(&c->rb_node, &rb_root);
+ if (compression)
+ account_unique_capacity(i->offset, unique_capacity, zc);
add:
add_item(c, i);
}
static void insert_chunks(struct item *items, unsigned int nitems,
- uint64_t *ndupes)
+ uint64_t *ndupes, uint64_t *unique_capacity,
+ struct zlib_ctrl *zc)
{
int i;
@@ -248,7 +289,7 @@ static void insert_chunks(struct item *items, unsigned int nitems,
r = bloom_set(bloom, items[i].hash, s);
*ndupes += r;
} else
- insert_chunk(&items[i]);
+ insert_chunk(&items[i], unique_capacity, zc);
}
fio_sem_up(rb_lock);
@@ -277,11 +318,13 @@ static int do_work(struct worker_thread *thread, void *buf)
off_t offset;
int nitems = 0;
uint64_t ndupes = 0;
+ uint64_t unique_capacity = 0;
struct item *items;
offset = thread->cur_offset;
- nblocks = read_blocks(thread->fd, buf, offset, min(thread->size, (uint64_t)chunk_size));
+ nblocks = read_blocks(thread->fd, buf, offset,
+ min(thread->size, (uint64_t) chunk_size));
if (!nblocks)
return 1;
@@ -296,20 +339,39 @@ static int do_work(struct worker_thread *thread, void *buf)
nitems++;
}
- insert_chunks(items, nitems, &ndupes);
+ insert_chunks(items, nitems, &ndupes, &unique_capacity, &thread->zc);
free(items);
thread->items += nitems;
thread->dupes += ndupes;
+ thread->unique_capacity += unique_capacity;
return 0;
}
+static void thread_init_zlib_control(struct worker_thread *thread)
+{
+ size_t sz;
+
+ z_stream *stream = &thread->zc.stream;
+ stream->zalloc = Z_NULL;
+ stream->zfree = Z_NULL;
+ stream->opaque = Z_NULL;
+
+ if (deflateInit(stream, Z_DEFAULT_COMPRESSION) != Z_OK)
+ return;
+
+ thread->zc.buf_in = fio_memalign(blocksize, blocksize, false);
+ sz = deflateBound(stream, blocksize);
+ thread->zc.buf_out = fio_memalign(blocksize, sz, false);
+}
+
static void *thread_fn(void *data)
{
struct worker_thread *thread = data;
void *buf;
buf = fio_memalign(blocksize, chunk_size, false);
+ thread_init_zlib_control(thread);
do {
if (get_work(&thread->cur_offset, &thread->size)) {
@@ -362,15 +424,17 @@ static void show_progress(struct worker_thread *threads, unsigned long total)
printf("%3.2f%% done (%luKiB/sec)\r", perc, this_items);
last_nitems = nitems;
fio_gettime(&last_tv, NULL);
- } else
+ } else {
printf("%3.2f%% done\r", perc);
+ }
fflush(stdout);
usleep(250000);
};
}
static int run_dedupe_threads(struct fio_file *f, uint64_t dev_size,
- uint64_t *nextents, uint64_t *nchunks)
+ uint64_t *nextents, uint64_t *nchunks,
+ uint64_t *unique_capacity)
{
struct worker_thread *threads;
unsigned long nitems, total_items;
@@ -398,11 +462,13 @@ static int run_dedupe_threads(struct fio_file *f, uint64_t dev_size,
nitems = 0;
*nextents = 0;
*nchunks = 1;
+ *unique_capacity = 0;
for (i = 0; i < num_threads; i++) {
void *ret;
pthread_join(threads[i].thread, &ret);
nitems += threads[i].items;
*nchunks += threads[i].dupes;
+ *unique_capacity += threads[i].unique_capacity;
}
printf("Threads(%u): %lu items processed\n", num_threads, nitems);
@@ -416,7 +482,7 @@ static int run_dedupe_threads(struct fio_file *f, uint64_t dev_size,
}
static int dedupe_check(const char *filename, uint64_t *nextents,
- uint64_t *nchunks)
+ uint64_t *nchunks, uint64_t *unique_capacity)
{
uint64_t dev_size;
struct stat sb;
@@ -451,9 +517,11 @@ static int dedupe_check(const char *filename, uint64_t *nextents,
bloom = bloom_new(bloom_entries);
}
- printf("Will check <%s>, size <%llu>, using %u threads\n", filename, (unsigned long long) dev_size, num_threads);
+ printf("Will check <%s>, size <%llu>, using %u threads\n", filename,
+ (unsigned long long) dev_size, num_threads);
- return run_dedupe_threads(&file, dev_size, nextents, nchunks);
+ return run_dedupe_threads(&file, dev_size, nextents, nchunks,
+ unique_capacity);
err:
if (file.fd != -1)
close(file.fd);
@@ -466,18 +534,38 @@ static void show_chunk(struct chunk *c)
struct flist_head *n;
struct extent *e;
- printf("c hash %8x %8x %8x %8x, count %lu\n", c->hash[0], c->hash[1], c->hash[2], c->hash[3], (unsigned long) c->count);
+ printf("c hash %8x %8x %8x %8x, count %lu\n", c->hash[0], c->hash[1],
+ c->hash[2], c->hash[3], (unsigned long) c->count);
flist_for_each(n, &c->extent_list[0]) {
e = flist_entry(n, struct extent, list);
printf("\toffset %llu\n", (unsigned long long) e->offset);
}
}
-static void show_stat(uint64_t nextents, uint64_t nchunks, uint64_t ndupextents)
+static const char *capacity_unit[] = {"b","KB", "MB", "GB", "TB", "PB", "EB"};
+
+static uint64_t bytes_to_human_readable_unit(uint64_t n, const char **unit_out)
+{
+ uint8_t i = 0;
+
+ while (n >= 1024) {
+ i++;
+ n /= 1024;
+ }
+
+ *unit_out = capacity_unit[i];
+ return n;
+}
+
+static void show_stat(uint64_t nextents, uint64_t nchunks, uint64_t ndupextents,
+ uint64_t unique_capacity)
{
double perc, ratio;
+ const char *unit;
+ uint64_t uc_human;
- printf("Extents=%lu, Unique extents=%lu", (unsigned long) nextents, (unsigned long) nchunks);
+ printf("Extents=%lu, Unique extents=%lu", (unsigned long) nextents,
+ (unsigned long) nchunks);
if (!bloom)
printf(" Duplicated extents=%lu", (unsigned long) ndupextents);
printf("\n");
@@ -485,22 +573,29 @@ static void show_stat(uint64_t nextents, uint64_t nchunks, uint64_t ndupextents)
if (nchunks) {
ratio = (double) nextents / (double) nchunks;
printf("De-dupe ratio: 1:%3.2f\n", ratio - 1.0);
- } else
+ } else {
printf("De-dupe ratio: 1:infinite\n");
+ }
- if (ndupextents)
- printf("De-dupe working set at least: %3.2f%%\n", 100.0 * (double) ndupextents / (double) nextents);
+ if (ndupextents) {
+ printf("De-dupe working set at least: %3.2f%%\n",
+ 100.0 * (double) ndupextents / (double) nextents);
+ }
perc = 1.00 - ((double) nchunks / (double) nextents);
perc *= 100.0;
printf("Fio setting: dedupe_percentage=%u\n", (int) (perc + 0.50));
+
+ if (compression) {
+ uc_human = bytes_to_human_readable_unit(unique_capacity, &unit);
+ printf("Unique capacity %lu%s\n", (unsigned long) uc_human, unit);
+ }
}
static void iter_rb_tree(uint64_t *nextents, uint64_t *nchunks, uint64_t *ndupextents)
{
struct fio_rb_node *n;
-
*nchunks = *nextents = *ndupextents = 0;
n = rb_first(&rb_root);
@@ -532,18 +627,19 @@ static int usage(char *argv[])
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");
+ log_err("\t-C\tCalculate compressible size\n");
return 1;
}
int main(int argc, char *argv[])
{
- uint64_t nextents = 0, nchunks = 0, ndupextents = 0;
+ uint64_t nextents = 0, nchunks = 0, ndupextents = 0, unique_capacity;
int c, ret;
arch_init(argv);
debug_init();
- while ((c = getopt(argc, argv, "b:t:d:o:c:p:B:")) != -1) {
+ while ((c = getopt(argc, argv, "b:t:d:o:c:p:B:C:")) != -1) {
switch (c) {
case 'b':
blocksize = atoi(optarg);
@@ -566,13 +662,16 @@ int main(int argc, char *argv[])
case 'B':
use_bloom = atoi(optarg);
break;
+ case 'C':
+ compression = atoi(optarg);
+ break;
case '?':
default:
return usage(argv);
}
}
- if (collision_check || dump_output)
+ if (collision_check || dump_output || compression)
use_bloom = 0;
if (!num_threads)
@@ -586,13 +685,13 @@ int main(int argc, char *argv[])
rb_root = RB_ROOT;
rb_lock = fio_sem_init(FIO_SEM_UNLOCKED);
- ret = dedupe_check(argv[optind], &nextents, &nchunks);
+ ret = dedupe_check(argv[optind], &nextents, &nchunks, &unique_capacity);
if (!ret) {
if (!bloom)
iter_rb_tree(&nextents, &nchunks, &ndupextents);
- show_stat(nextents, nchunks, ndupextents);
+ show_stat(nextents, nchunks, ndupextents, unique_capacity);
}
fio_sem_remove(rb_lock);
diff --git a/t/io_uring.c b/t/io_uring.c
index 7bf215c7..a98f78fd 100644
--- a/t/io_uring.c
+++ b/t/io_uring.c
@@ -192,7 +192,7 @@ unsigned int calc_clat_percentiles(unsigned long *io_u_plat, unsigned long nr,
unsigned long *ovals = NULL;
bool is_last;
- *minv = -1ULL;
+ *minv = -1UL;
*maxv = 0;
ovals = malloc(len * sizeof(*ovals));
@@ -498,7 +498,7 @@ static void init_io(struct submitter *s, unsigned index)
sqe->off = offset;
sqe->user_data = (unsigned long) f->fileno;
if (stats && stats_running)
- sqe->user_data |= ((unsigned long)s->clock_index << 32);
+ sqe->user_data |= ((uint64_t)s->clock_index << 32);
}
static int prep_more_ios_uring(struct submitter *s, int max_ios)
