mempool: test performance with constant n
PATCH 1/1 mempool: test performance with constant n
"What gets measured gets done."
This patch adds mempool performance tests where the number of objects to put and get is constant at compile time, which may significantly improve the performance of these functions. [*]
Also, it is ensured that the array holding the object used for testing is cache line aligned, for maximum performance.
And finally, the following entries are added to the list of tests:
- Number of kept objects: 512
- Number of objects to get and to put: The number of pointers fitting into a cache line, i.e. 8 or 16
[*] Some example performance test (with cache) results:
get_bulk=4 put_bulk=4 keep=128 constant_n=false rate_persec=280480972 get_bulk=4 put_bulk=4 keep=128 constant_n=true rate_persec=622159462
get_bulk=8 put_bulk=8 keep=128 constant_n=false rate_persec=477967155 get_bulk=8 put_bulk=8 keep=128 constant_n=true rate_persec=917582643
get_bulk=32 put_bulk=32 keep=32 constant_n=false rate_persec=871248691 get_bulk=32 put_bulk=32 keep=32 constant_n=true rate_persec=1134021836
--- app/test/test_mempool_perf.c | 120 +++++++++++++++++++++-------------- 1 file changed, 74 insertions(+), 46 deletions(-)
app/test/test_mempool_perf.c +74 -46
@@ -1,5 +1,6 @@/* SPDX-License-Identifier: BSD-3-Clause * Copyright(c) 2010-2014 Intel Corporation +* Copyright(c) 2022 SmartShare Systems */ #include <string.h>
@@ -55,19 +56,24 @@* * - Bulk get from 1 to 32 * - Bulk put from 1 to 32 +* - Bulk get and put from 1 to 32, compile time constant * * - Number of kept objects (*n_keep*) * * - 32 * - 128 +* - 512 */ #define N 65536#define TIME_S 5#define MEMPOOL_ELT_SIZE 2048 -#define MAX_KEEP 128 +#define MAX_KEEP 512#define MEMPOOL_SIZE ((rte_lcore_count()*(MAX_KEEP+RTE_MEMPOOL_CACHE_MAX_SIZE))-1) +/* Number of pointers fitting into one cache line. */ +#define CACHE_LINE_BURST (RTE_CACHE_LINE_SIZE/sizeof(uintptr_t)) +#define LOG_ERR () printf("test failed at %s():%d\n", __func__, __LINE__)#define RET_ERR () do { \ LOG_ERR(); \
@@ -80,16 +86,16 @@ }while (0 ) static int use_external_cache; -static unsigned external_cache_size =RTE_MEMPOOL_CACHE_MAX_SIZE ; +static unsigned int external_cache_size =RTE_MEMPOOL_CACHE_MAX_SIZE ;static uint32_t synchro;/* number of objects in one bulk operation (get or put) */ -static unsigned n_get_bulk; -static unsigned n_put_bulk; +static int n_get_bulk; +static int n_put_bulk;/* number of objects retrieved from mempool before putting them back */ -static unsigned n_keep; +static int n_keep;/* number of enqueues / dequeues */ struct mempool_test_stats {
@@ -104,20 +110,43 @@ static struct mempool_test_stats stats[RTE_MAX_LCORE];*/ static void my_obj_init(struct rte_mempool *mp, __rte_unused void *arg, -void *obj,unsigned i) +void *obj,unsigned int i) {uint32_t *objnum = obj;memset (obj,0 , mp->elt_size); *objnum = i; } +#define test_loop (x_keep, x_get_bulk, x_put_bulk) \ + for (i = 0; likely(i < (N/x_keep)); i++) { \ +/* get x_keep objects by bulk of x_get_bulk */ \ +for (idx =0 ; idx < x_keep; idx += x_get_bulk) {\ + ret =rte_mempool_generic_get (mp, \ + &obj_table[idx], \ + x_get_bulk, \ + cache); \ +if (unlikely (ret <0 )) { \ +rte_mempool_dump (stdout, mp); \ +GOTO_ERR (ret, out); \ + } \ + } \ + \ +/* put the objects back by bulk of x_put_bulk */ \ +for (idx =0 ; idx < x_keep; idx += x_put_bulk) {\ +rte_mempool_generic_put (mp, \ + &obj_table[idx], \ + x_put_bulk, \ + cache); \ + } \ + } +static int per_lcore_mempool_test (void *arg) { -void *obj_table[MAX_KEEP ]; -unsigned i, idx; + void *obj_table[MAX_KEEP] __rte_cache_aligned; +int i, idx;struct rte_mempool *mp = arg; -unsigned lcore_id =rte_lcore_id (); +unsigned int lcore_id =rte_lcore_id ();int ret =0 ;uint64_t start_cycles, end_cycles;uint64_t time_diff =0 , hz =rte_get_timer_hz ();
@@ -139,6 +168,9 @@ per_lcore_mempool_test(void *arg)GOTO_ERR (ret, out);if (((n_keep / n_put_bulk) * n_put_bulk) != n_keep)GOTO_ERR (ret, out); +/* for constant n, n_get_bulk and n_put_bulk must be the same */ +if (n_get_bulk <0 && n_put_bulk != n_get_bulk) +GOTO_ERR (ret, out); stats[lcore_id].enq_count =0 ;
@@ -149,30 +181,18 @@ per_lcore_mempool_test(void *arg) start_cycles =rte_get_timer_cycles ();while (time_diff/hz <TIME_S ) { -for (i =0 ;likely (i < (N /n_keep)); i++) { -/* get n_keep objects by bulk of n_bulk */ - idx =0 ; -while (idx < n_keep) { - ret =rte_mempool_generic_get (mp, - &obj_table[idx], - n_get_bulk, - cache); -if (unlikely (ret <0 )) { -rte_mempool_dump (stdout, mp); -/* in this case, objects are lost... */ -GOTO_ERR (ret, out); - } - idx += n_get_bulk; - } - -/* put the objects back */ - idx =0 ; -while (idx < n_keep) { -rte_mempool_generic_put (mp, &obj_table[idx], - n_put_bulk, - cache); - idx += n_put_bulk; - } +if (n_get_bulk >0 ) { +test_loop (n_keep, n_get_bulk, n_put_bulk); + }else if (n_get_bulk ==-1 ) { +test_loop (-n_keep,1 ,1 ); + }else if (n_get_bulk ==-4 ) { +test_loop (-n_keep,4 ,4 ); + }else if (n_get_bulk == -(int )CACHE_LINE_BURST ) { +test_loop (-n_keep,CACHE_LINE_BURST ,CACHE_LINE_BURST ); + }else if (n_get_bulk ==-32 ) { +test_loop (-n_keep,32 ,32 ); + }else { +GOTO_ERR (ret, out); } end_cycles =rte_get_timer_cycles (); time_diff = end_cycles - start_cycles;
@@ -192,10 +212,10 @@ per_lcore_mempool_test(void *arg)static int launch_cores (struct rte_mempool *mp,unsigned int cores) { -unsigned lcore_id; +unsigned int lcore_id;uint64_t rate;int ret; -unsigned cores_save = cores; +unsigned int cores_save = cores;__atomic_store_n (&synchro,0 , __ATOMIC_RELAXED);
@@ -203,10 +223,14 @@ launch_cores(struct rte_mempool *mp, unsigned int cores)memset (stats,0 ,sizeof (stats));printf ("mempool_autotest cache=%u cores=%u n_get_bulk=%u " -"n_put_bulk=%u n_keep=%u " , +"n_put_bulk=%u n_keep=%u constant_n=%s " , use_external_cache ? - external_cache_size : (unsigned ) mp->cache_size, - cores, n_get_bulk, n_put_bulk, n_keep); + external_cache_size : (unsigned int ) mp->cache_size, + cores, + n_get_bulk >0 ? n_get_bulk : -n_get_bulk, + n_put_bulk >0 ? n_put_bulk : -n_put_bulk, + n_keep >0 ? n_keep : -n_keep, + n_get_bulk >0 ?"false" :"true" );if (rte_mempool_avail_count (mp) !=MEMPOOL_SIZE ) {printf ("mempool is not full\n" );
@@ -253,12 +277,13 @@ launch_cores(struct rte_mempool *mp, unsigned int cores)static int do_one_mempool_test (struct rte_mempool *mp,unsigned int cores) { -unsigned bulk_tab_get[] = {1 ,4 ,32 ,0 }; -unsigned bulk_tab_put[] = {1 ,4 ,32 ,0 }; -unsigned keep_tab[] = {32 ,128 ,0 }; -unsigned *get_bulk_ptr; -unsigned *put_bulk_ptr; -unsigned *keep_ptr; +/* Negative n_get_bulk values represent constants in the test. */ +int bulk_tab_get[] = {1 ,4 ,CACHE_LINE_BURST ,32 ,-1 ,-4 , -(int )CACHE_LINE_BURST ,-32 ,0 }; +int bulk_tab_put[] = {1 ,4 ,CACHE_LINE_BURST ,32 ,0 }; +int keep_tab[] = {32 ,128 ,512 ,0 }; +int *get_bulk_ptr; +int *put_bulk_ptr; +int *keep_ptr;int ret;for (get_bulk_ptr = bulk_tab_get; *get_bulk_ptr; get_bulk_ptr++) {
@@ -266,13 +291,16 @@ do_one_mempool_test(struct rte_mempool *mp, unsigned int cores)for (keep_ptr = keep_tab; *keep_ptr; keep_ptr++) { n_get_bulk = *get_bulk_ptr; - n_put_bulk = *put_bulk_ptr; - n_keep = *keep_ptr; + n_put_bulk = n_get_bulk >0 ? *put_bulk_ptr : n_get_bulk; + n_keep = n_get_bulk >0 ? *keep_ptr : -*keep_ptr; ret =launch_cores (mp, cores);if (ret <0 )return -1 ; } +/* Skip the put loop when n is constant */ +if (n_get_bulk <0 ) +break ; } }return 0 ;
Signature · 1 line
2.17.1
Hi Morten,
Thank you for enhancing the mempool test. Please see some comments below.
Quoted 54 lines · Morten Brørup
"What gets measured gets done."
This patch adds mempool performance tests where the number of objects to put and get is constant at compile time, which may significantly improve the performance of these functions. [*]
Also, it is ensured that the array holding the object used for testing is cache line aligned, for maximum performance.
And finally, the following entries are added to the list of tests:
- Number of kept objects: 512
- Number of objects to get and to put: The number of pointers fitting into a cache line, i.e. 8 or 16[*] Some example performance test (with cache) results:
get_bulk=4 put_bulk=4 keep=128 constant_n=false rate_persec=280480972 get_bulk=4 put_bulk=4 keep=128 constant_n=true rate_persec=622159462
get_bulk=8 put_bulk=8 keep=128 constant_n=false rate_persec=477967155 get_bulk=8 put_bulk=8 keep=128 constant_n=true rate_persec=917582643
get_bulk=32 put_bulk=32 keep=32 constant_n=false rate_persec=871248691 get_bulk=32 put_bulk=32 keep=32 constant_n=true rate_persec=1134021836
Signed-off-by Morten Brørup <[email protected]>--- app/test/test_mempool_perf.c | 120 +++++++++++++++++++++-------------- 1 file changed, 74 insertions(+), 46 deletions(-)app/test/test_mempool_perf.c +3
-0@@ -1,5 +1,6 @@/* SPDX-License-Identifier: BSD-3-Clause * Copyright(c) 2010-2014 Intel Corporation +* Copyright(c) 2022 SmartShare Systems */ #include <string.h> @@ -55,19 +56,24 @@* * - Bulk get from 1 to 32 * - Bulk put from 1 to 32 +* - Bulk get and put from 1 to 32, compile time constant * * - Number of kept objects (*n_keep*) * * - 32 * - 128 +* - 512 */
app/test/test_mempool_perf.c line 76
Quoted 10 lines
app/test/test_mempool_perf.c +4
-1#define N 65536#define TIME_S 5#define MEMPOOL_ELT_SIZE 2048 -#define MAX_KEEP 128 +#define MAX_KEEP 512#define MEMPOOL_SIZE ((rte_lcore_count()*(MAX_KEEP+RTE_MEMPOOL_CACHE_MAX_SIZE))-1) +/* Number of pointers fitting into one cache line. */ +#define CACHE_LINE_BURST (RTE_CACHE_LINE_SIZE/sizeof(uintptr_t)) +
nit: I think it's better to follow the coding rules and add a space around the '/', even if I can see the line right above does not follow this convention
Quoted 49 lines
app/test/test_mempool_perf.c +18
-5#define LOG_ERR () printf("test failed at %s():%d\n", __func__, __LINE__)#define RET_ERR () do { \ LOG_ERR(); \@@ -80,16 +86,16 @@ }while (0 ) static int use_external_cache; -static unsigned external_cache_size =RTE_MEMPOOL_CACHE_MAX_SIZE ; +static unsigned int external_cache_size =RTE_MEMPOOL_CACHE_MAX_SIZE ;static uint32_t synchro;/* number of objects in one bulk operation (get or put) */ -static unsigned n_get_bulk; -static unsigned n_put_bulk; +static int n_get_bulk; +static int n_put_bulk;/* number of objects retrieved from mempool before putting them back */ -static unsigned n_keep; +static int n_keep;/* number of enqueues / dequeues */ struct mempool_test_stats {@@ -104,20 +110,43 @@ static struct mempool_test_stats stats[RTE_MAX_LCORE];*/ static void my_obj_init(struct rte_mempool *mp, __rte_unused void *arg, -void *obj,unsigned i) +void *obj,unsigned int i) {uint32_t *objnum = obj;memset (obj,0 , mp->elt_size); *objnum = i; } +#define test_loop (x_keep, x_get_bulk, x_put_bulk) \ + for (i = 0; likely(i < (N/x_keep)); i++) { \ +/* get x_keep objects by bulk of x_get_bulk */ \ +for (idx =0 ; idx < x_keep; idx += x_get_bulk) {\ + ret =rte_mempool_generic_get (mp, \ + &obj_table[idx], \ + x_get_bulk, \ + cache); \ +if (unlikely (ret <0 )) { \ +rte_mempool_dump (stdout, mp); \ +GOTO_ERR (ret, out); \ + } \ + } \
app/test/test_mempool_perf.c static struct mempool_test_stats stats[RTE_MAX_LCORE]; · line 142
Quoted 10 lines
app/test/test_mempool_perf.c +10
-0+ \ +/* put the objects back by bulk of x_put_bulk */ \ +for (idx =0 ; idx < x_keep; idx += x_put_bulk) {\ +rte_mempool_generic_put (mp, \ + &obj_table[idx], \ + x_put_bulk, \ + cache); \ + } \ + } +
I think a static __rte_always_inline function would do the job and is clearer. Something like this:
static __rte_always_inline int
test_loop(struct rte_mempool *mp, struct rte_mempool_cache *cache,
unsigned int x_keep, unsigned int x_get_bulk, unsigned int x_put_bulk)
{
void *obj_table[MAX_KEEP] __rte_cache_aligned;
unsigned int idx;
unsigned int i;
int ret;
for (i = 0; likely(i < (N / x_keep)); i++) {
/* get x_keep objects by bulk of x_get_bulk */
for (idx = 0; idx < x_keep; idx += x_get_bulk) {
ret = rte_mempool_generic_get(mp,
&obj_table[idx],
x_get_bulk,
cache);
if (unlikely(ret < 0)) {
rte_mempool_dump(stdout, mp);
return ret;
}
}
/* put the objects back by bulk of x_put_bulk */
for (idx = 0; idx < x_keep; idx += x_put_bulk) {
rte_mempool_generic_put(mp,
&obj_table[idx],
x_put_bulk,
cache);
}
}
return 0;
}Quoted 56 lines
app/test/test_mempool_perf.c +11
-27static int per_lcore_mempool_test (void *arg) { -void *obj_table[MAX_KEEP ]; -unsigned i, idx; + void *obj_table[MAX_KEEP] __rte_cache_aligned; +int i, idx;struct rte_mempool *mp = arg; -unsigned lcore_id =rte_lcore_id (); +unsigned int lcore_id =rte_lcore_id ();int ret =0 ;uint64_t start_cycles, end_cycles;uint64_t time_diff =0 , hz =rte_get_timer_hz ();@@ -139,6 +168,9 @@ per_lcore_mempool_test(void *arg)GOTO_ERR (ret, out);if (((n_keep / n_put_bulk) * n_put_bulk) != n_keep)GOTO_ERR (ret, out); +/* for constant n, n_get_bulk and n_put_bulk must be the same */ +if (n_get_bulk <0 && n_put_bulk != n_get_bulk) +GOTO_ERR (ret, out); stats[lcore_id].enq_count =0 ;@@ -149,30 +181,18 @@ per_lcore_mempool_test(void *arg) start_cycles =rte_get_timer_cycles ();while (time_diff/hz <TIME_S ) { -for (i =0 ;likely (i < (N /n_keep)); i++) { -/* get n_keep objects by bulk of n_bulk */ - idx =0 ; -while (idx < n_keep) { - ret =rte_mempool_generic_get (mp, - &obj_table[idx], - n_get_bulk, - cache); -if (unlikely (ret <0 )) { -rte_mempool_dump (stdout, mp); -/* in this case, objects are lost... */ -GOTO_ERR (ret, out); - } - idx += n_get_bulk; - } - -/* put the objects back */ - idx =0 ; -while (idx < n_keep) { -rte_mempool_generic_put (mp, &obj_table[idx], - n_put_bulk, - cache); - idx += n_put_bulk; - } +if (n_get_bulk >0 ) { +test_loop (n_keep, n_get_bulk, n_put_bulk); + }else if (n_get_bulk ==-1 ) { +test_loop (-n_keep,1 ,1 ); + }else if (n_get_bulk ==-4 ) {
app/test/test_mempool_perf.c per_lcore_mempool_test(void *arg) · line 198
Quoted 10 lines
app/test/test_mempool_perf.c +7
-0+test_loop (-n_keep,4 ,4 ); + }else if (n_get_bulk == -(int )CACHE_LINE_BURST ) { +test_loop (-n_keep,CACHE_LINE_BURST ,CACHE_LINE_BURST ); + }else if (n_get_bulk ==-32 ) { +test_loop (-n_keep,32 ,32 ); + }else { +GOTO_ERR (ret, out); } end_cycles =rte_get_timer_cycles (); time_diff = end_cycles - start_cycles;
I'm not convinced that having negative values to mean "constant" is clear. I'd prefer to have another global variable "use_constant_values", which would give something like this:
if (!use_constant_values) ret = test_loop(mp, cache, n_keep, n_get_bulk, n_put_bulk); else if (n_get_bulk == 1) ret = test_loop(mp, cache, n_keep, 1, 1); else if (n_get_bulk == 4) ret = test_loop(mp, cache, n_keep, 4, 4); else if (n_get_bulk == CACHE_LINE_BURST) ret = test_loop(mp, cache, n_keep, CACHE_LINE_BURST, CACHE_LINE_BURST); else if (n_get_bulk == 32) ret = test_loop(mp, cache, n_keep, 32, 32); else ret = -1; if (ret < 0) GOTO_ERR(ret, out);
Quoted 19 lines
app/test/test_mempool_perf.c +3
-3@@ -192,10 +212,10 @@ per_lcore_mempool_test(void *arg)static int launch_cores (struct rte_mempool *mp,unsigned int cores) { -unsigned lcore_id; +unsigned int lcore_id;uint64_t rate;int ret; -unsigned cores_save = cores; +unsigned int cores_save = cores;__atomic_store_n (&synchro,0 , __ATOMIC_RELAXED);@@ -203,10 +223,14 @@ launch_cores(struct rte_mempool *mp, unsigned int cores)memset (stats,0 ,sizeof (stats)); printf("mempool_autotest cache=%u cores=%u n_get_bulk=%u " - "n_put_bulk=%u n_keep=%u ", + "n_put_bulk=%u n_keep=%u constant_n=%s ",
app/test/test_mempool_perf.c launch_cores(struct rte_mempool *mp, unsigned int cores) · line 234
Quoted 10 lines
app/test/test_mempool_perf.c +6
-2use_external_cache ? - external_cache_size : (unsigned ) mp->cache_size, - cores, n_get_bulk, n_put_bulk, n_keep); + external_cache_size : (unsigned int ) mp->cache_size, + cores, + n_get_bulk >0 ? n_get_bulk : -n_get_bulk, + n_put_bulk >0 ? n_put_bulk : -n_put_bulk, + n_keep >0 ? n_keep : -n_keep, + n_get_bulk > 0 ? "false" : "true");
This would become much simpler with this new variable.
Quoted 11 lines
app/test/test_mempool_perf.c +0
-5if (rte_mempool_avail_count (mp) !=MEMPOOL_SIZE ) {printf ("mempool is not full\n" );@@ -253,12 +277,13 @@ launch_cores(struct rte_mempool *mp, unsigned int cores)static int do_one_mempool_test (struct rte_mempool *mp,unsigned int cores) { -unsigned bulk_tab_get[] = {1 ,4 ,32 ,0 }; -unsigned bulk_tab_put[] = {1 ,4 ,32 ,0 }; -unsigned keep_tab[] = {32 ,128 ,0 }; -unsigned *get_bulk_ptr; -unsigned *put_bulk_ptr;
app/test/test_mempool_perf.c launch_cores(struct rte_mempool *mp, unsigned int cores) · line 288
Quoted 10 lines
app/test/test_mempool_perf.c +7
-1-unsigned *keep_ptr; +/* Negative n_get_bulk values represent constants in the test. */ +int bulk_tab_get[] = {1 ,4 ,CACHE_LINE_BURST ,32 ,-1 ,-4 , -(int )CACHE_LINE_BURST ,-32 ,0 }; +int bulk_tab_put[] = {1 ,4 ,CACHE_LINE_BURST ,32 ,0 }; +int keep_tab[] = {32 ,128 ,512 ,0 }; +int *get_bulk_ptr; +int *put_bulk_ptr; +int *keep_ptr;int ret;
Same here, changes would be minimal.
app/test/test_mempool_perf.c do_one_mempool_test(struct rte_mempool *mp, unsigned int cores) · line 299
Quoted 13 lines
app/test/test_mempool_perf.c +2
-2for (get_bulk_ptr = bulk_tab_get; *get_bulk_ptr; get_bulk_ptr++) {@@ -266,13 +291,16 @@ do_one_mempool_test(struct rte_mempool *mp, unsigned int cores)for (keep_ptr = keep_tab; *keep_ptr; keep_ptr++) { n_get_bulk = *get_bulk_ptr; - n_put_bulk = *put_bulk_ptr; - n_keep = *keep_ptr; + n_put_bulk = n_get_bulk >0 ? *put_bulk_ptr : n_get_bulk; + n_keep = n_get_bulk >0 ? *keep_ptr : -*keep_ptr; ret =launch_cores (mp, cores);if (ret <0 )return -1 ;
No change would be required above (except use_constant_values = 0). And below, we could simply add instead:
/* replay test with constant values */
if (n_get_bulk == n_put_bulk) {
use_constant_values = 1;
ret = launch_cores(mp, cores);
if (ret < 0)
return -1;
}If you are ok with the proposed changes, I can directly submit a v2, since I already made them locally.
Thanks,
Olivier
Quoted 372 lines
From: Olivier Matz [mailto:[email protected]]
Sent: Monday, 24 January 2022 11.26Hi Morten,
Thank you for enhancing the mempool test. Please see some comments below.
Quoted 3 lines · Morten Brørup
"What gets measured gets done."
This patch adds mempool performance tests where the number of objects
to
Quoted 1 line
put and get is constant at compile time, which may significantly
improve
Quoted 3 lines
the performance of these functions. [*]
Also, it is ensured that the array holding the object used for
testing
Quoted 9 lines
is cache line aligned, for maximum performance.
And finally, the following entries are added to the list of tests:
- Number of kept objects: 512
- Number of objects to get and to put: The number of pointers fitting into a cache line, i.e. 8 or 16[*] Some example performance test (with cache) results:
Quoted 7 lines
get_bulk=4 put_bulk=4 keep=128 constant_n=false rate_persec=280480972 get_bulk=4 put_bulk=4 keep=128 constant_n=true rate_persec=622159462
get_bulk=8 put_bulk=8 keep=128 constant_n=false rate_persec=477967155 get_bulk=8 put_bulk=8 keep=128 constant_n=true rate_persec=917582643
get_bulk=32 put_bulk=32 keep=32 constant_n=false
rate_persec=871248691
Quoted 1 line
get_bulk=32 put_bulk=32 keep=32 constant_n=true
rate_persec=1134021836
Quoted 4 lines
Signed-off-by Morten Brørup <[email protected]>--- app/test/test_mempool_perf.c | 120 +++++++++++++++++++++--------------
Review
Quoted 3 lines
1 file changed, 74 insertions(+), 46 deletions(-)(quoted diff) +0
-0
b/app/test/test_mempool_perf.cQuoted 19 lines
index
87ad251367..ffefe934d5100644app/test/test_mempool_perf.c +2
-0@@ -1,5 +1,6 @@/* SPDX-License-Identifier: BSD-3-Clause * Copyright(c) 2010-2014 Intel Corporation +* Copyright(c) 2022 SmartShare Systems */ #include <string.h> @@ -55,19 +56,24 @@* * - Bulk get from 1 to 32 * - Bulk put from 1 to 32 +* - Bulk get and put from 1 to 32, compile time constant * * - Number of kept objects (*n_keep*) * * - 32 app/test/test_mempool_perf.c line 72
Quoted 10 lines
app/test/test_mempool_perf.c +2
-1* - 128 +* - 512 */ #define N 65536#define TIME_S 5#define MEMPOOL_ELT_SIZE 2048 -#define MAX_KEEP 128 +#define MAX_KEEP 512#define MEMPOOL_SIZE ((rte_lcore_count()*(MAX_KEEP+RTE_MEMPOOL_CACHE_MAX_SIZE))-1)
app/test/test_mempool_perf.c line 76
Quoted 4 lines
app/test/test_mempool_perf.c +3
-0+/* Number of pointers fitting into one cache line. */ +#define CACHE_LINE_BURST (RTE_CACHE_LINE_SIZE/sizeof(uintptr_t)) +nit: I think it's better to follow the coding rules and add a space around the
'/', even if I can see the line right above does not follow this conventionapp/test/test_mempool_perf.c line 76
Quoted 1 line
app/test/test_mempool_perf.c +0
-0#define LOG_ERR () printf("test failed at %s():%d\n", __func__,
__LINE__)app/test/test_mempool_perf.c line 89
Quoted 8 lines
app/test/test_mempool_perf.c +1
-1#define RET_ERR () do { \ LOG_ERR(); \@@ -80,16 +86,16 @@ }while (0 ) static int use_external_cache; -static unsigned external_cache_size =RTE_MEMPOOL_CACHE_MAX_SIZE ; +static unsigned int external_cache_size =
RTE_MEMPOOL_CACHE_MAX_SIZE;app/test/test_mempool_perf.c
Quoted 10 lines
app/test/test_mempool_perf.c +2
-2static uint32_t synchro;/* number of objects in one bulk operation (get or put) */ -static unsigned n_get_bulk; -static unsigned n_put_bulk; +static int n_get_bulk; +static int n_put_bulk;/* number of objects retrieved from mempool before putting them back */
app/test/test_mempool_perf.c static struct mempool_test_stats
Quoted 6 lines
app/test/test_mempool_perf.c +1
-1-static unsigned n_keep; +static int n_keep;/* number of enqueues / dequeues */ struct mempool_test_stats {@@ -104,20 +110,43 @@ static struct mempool_test_statsstats[RTE_MAX_LCORE];
Quoted 24 lines
app/test/test_mempool_perf.c +14
-1*/ static void my_obj_init(struct rte_mempool *mp, __rte_unused void *arg, -void *obj,unsigned i) +void *obj,unsigned int i) {uint32_t *objnum = obj;memset (obj,0 , mp->elt_size); *objnum = i; } +#define test_loop (x_keep, x_get_bulk, x_put_bulk) \ + for (i = 0; likely(i < (N/x_keep)); i++) { \ +/* get x_keep objects by bulk of x_get_bulk */ \ +for (idx =0 ; idx < x_keep; idx += x_get_bulk) {\ + ret =rte_mempool_generic_get (mp, \ + &obj_table[idx], \ + x_get_bulk, \ + cache); \ +if (unlikely (ret <0 )) { \ +rte_mempool_dump (stdout, mp); \ +GOTO_ERR (ret, out); \ + } \ + } \app/test/test_mempool_perf.c static struct mempool_test_stats stats[RTE_MAX_LCORE]; · line 142
Quoted 10 lines
app/test/test_mempool_perf.c +10
-0+ \ +/* put the objects back by bulk of x_put_bulk */ \ +for (idx =0 ; idx < x_keep; idx += x_put_bulk) {\ +rte_mempool_generic_put (mp, \ + &obj_table[idx], \ + x_put_bulk, \ + cache); \ + } \ + } +I think a static
__rte_always_inlinefunction would do the job and is clearer. Something like this:static
__rte_always_inlineint
test_loop(structrte_mempool*mp, structrte_mempool_cache*cache,
unsigned intx_keep, unsigned intx_get_bulk, unsigned intx_put_bulk){ void *obj_table[MAX_KEEP] __rte_cache_aligned; unsigned int idx; unsigned int i; int ret; for (i = 0; likely(i < (N / x_keep)); i++) { /* get x_keep objects by bulk of x_get_bulk */ for (idx = 0; idx < x_keep; idx += x_get_bulk) { ret = rte_mempool_generic_get(mp, &obj_table[idx], x_get_bulk, cache); if (unlikely(ret < 0)) { rte_mempool_dump(stdout, mp); return ret; } } /* put the objects back by bulk of x_put_bulk */ for (idx = 0; idx < x_keep; idx += x_put_bulk) { rte_mempool_generic_put(mp, &obj_table[idx], x_put_bulk, cache); } } return 0; }Quoted 49 lines
app/test/test_mempool_perf.c +6
-25static int per_lcore_mempool_test (void *arg) { -void *obj_table[MAX_KEEP ]; -unsigned i, idx; + void *obj_table[MAX_KEEP] __rte_cache_aligned; +int i, idx;struct rte_mempool *mp = arg; -unsigned lcore_id =rte_lcore_id (); +unsigned int lcore_id =rte_lcore_id ();int ret =0 ;uint64_t start_cycles, end_cycles;uint64_t time_diff =0 , hz =rte_get_timer_hz ();@@ -139,6 +168,9 @@ per_lcore_mempool_test(void *arg)GOTO_ERR (ret, out);if (((n_keep / n_put_bulk) * n_put_bulk) != n_keep)GOTO_ERR (ret, out); +/* for constant n, n_get_bulk and n_put_bulk must be the same */ +if (n_get_bulk <0 && n_put_bulk != n_get_bulk) +GOTO_ERR (ret, out); stats[lcore_id].enq_count =0 ;@@ -149,30 +181,18 @@ per_lcore_mempool_test(void *arg) start_cycles =rte_get_timer_cycles ();while (time_diff/hz <TIME_S ) { -for (i =0 ;likely (i < (N /n_keep)); i++) { -/* get n_keep objects by bulk of n_bulk */ - idx =0 ; -while (idx < n_keep) { - ret =rte_mempool_generic_get (mp, - &obj_table[idx], - n_get_bulk, - cache); -if (unlikely (ret <0 )) { -rte_mempool_dump (stdout, mp); -/* in this case, objects are lost... */ -GOTO_ERR (ret, out); - } - idx += n_get_bulk; - } - -/* put the objects back */ - idx =0 ; -while (idx < n_keep) { -rte_mempool_generic_put (mp, &obj_table[idx], - n_put_bulk, - cache);app/test/test_mempool_perf.c per_lcore_mempool_test(void *arg) · line 191
Quoted 10 lines
app/test/test_mempool_perf.c +8
-2- idx += n_put_bulk; - } +if (n_get_bulk >0 ) { +test_loop (n_keep, n_get_bulk, n_put_bulk); + }else if (n_get_bulk ==-1 ) { +test_loop (-n_keep,1 ,1 ); + }else if (n_get_bulk ==-4 ) { +test_loop (-n_keep,4 ,4 ); + }else if (n_get_bulk == -(int )CACHE_LINE_BURST ) { + test_loop(-n_keep, CACHE_LINE_BURST,
CACHE_LINE_BURST);app/test/test_mempool_perf.c per_lcore_mempool_test(void *arg) · line 198
Quoted 7 lines
app/test/test_mempool_perf.c +4
-0+ }else if (n_get_bulk ==-32 ) { +test_loop (-n_keep,32 ,32 ); + }else { +GOTO_ERR (ret, out); } end_cycles =rte_get_timer_cycles (); time_diff = end_cycles - start_cycles;I'm not convinced that having negative values to mean "constant" is clear. I'd prefer to have another global variable "
use_constant_values",
which would give something like this:if (!use_constant_values)
ret = test_loop(mp, cache,n_keep,n_get_bulk,n_put_bulk); else if (n_get_bulk == 1) ret = test_loop(mp, cache, n_keep, 1, 1); else if (n_get_bulk == 4) ret = test_loop(mp, cache, n_keep, 4, 4); else if (n_get_bulk == CACHE_LINE_BURST) ret = test_loop(mp, cache, n_keep, CACHE_LINE_BURST, CACHE_LINE_BURST); else if (n_get_bulk == 32) ret = test_loop(mp, cache, n_keep, 32, 32); else ret = -1; if (ret < 0) GOTO_ERR(ret, out);app/test/test_mempool_perf.c launch_cores(struct rte_mempool *mp, unsigned
Quoted 14 lines
app/test/test_mempool_perf.c +2
-2@@ -192,10 +212,10 @@ per_lcore_mempool_test(void *arg)static int launch_cores (struct rte_mempool *mp,unsigned int cores) { -unsigned lcore_id; +unsigned int lcore_id;uint64_t rate;int ret; -unsigned cores_save = cores; +unsigned int cores_save = cores;__atomic_store_n (&synchro,0 , __ATOMIC_RELAXED);@@ -203,10 +223,14 @@ launch_cores(struct rte_mempool *mp, unsignedint cores)
Quoted 5 lines
app/test/test_mempool_perf.c +1
-1memset (stats,0 ,sizeof (stats)); printf("mempool_autotest cache=%u cores=%u n_get_bulk=%u " - "n_put_bulk=%u n_keep=%u ", + "n_put_bulk=%u n_keep=%u constant_n=%s ",app/test/test_mempool_perf.c launch_cores(struct rte_mempool *mp, unsigned int cores) · line 234
Quoted 10 lines
app/test/test_mempool_perf.c +6
-2use_external_cache ? - external_cache_size : (unsigned ) mp->cache_size, - cores, n_get_bulk, n_put_bulk, n_keep); + external_cache_size : (unsigned int ) mp->cache_size, + cores, + n_get_bulk >0 ? n_get_bulk : -n_get_bulk, + n_put_bulk >0 ? n_put_bulk : -n_put_bulk, + n_keep >0 ? n_keep : -n_keep, + n_get_bulk > 0 ? "false" : "true");This would become much simpler with this new variable.
app/test/test_mempool_perf.c launch_cores(struct rte_mempool *mp, unsigned
Quoted 3 lines
app/test/test_mempool_perf.c +0
-0if (rte_mempool_avail_count (mp) !=MEMPOOL_SIZE ) {printf ("mempool is not full\n" );@@ -253,12 +277,13 @@ launch_cores(struct rte_mempool *mp, unsignedint cores)
app/test/test_mempool_perf.c launch_cores(struct rte_mempool *mp, unsigned int cores)
Quoted 11 lines
app/test/test_mempool_perf.c +2
-6static int do_one_mempool_test(struct rte_mempool *mp, unsigned int cores) { -unsigned bulk_tab_get[] = {1 ,4 ,32 ,0 }; -unsigned bulk_tab_put[] = {1 ,4 ,32 ,0 }; -unsigned keep_tab[] = {32 ,128 ,0 }; -unsigned *get_bulk_ptr; -unsigned *put_bulk_ptr; -unsigned *keep_ptr; + /* Negative n_get_bulk values represent constants in the test. */ + int bulk_tab_get[] = { 1, 4, CACHE_LINE_BURST, 32, -1, -4, -(int)CACHE_LINE_BURST, -32, 0 };
app/test/test_mempool_perf.c launch_cores(struct rte_mempool *mp, unsigned int cores) · line 288
Quoted 7 lines
app/test/test_mempool_perf.c +5
-0+int bulk_tab_put[] = {1 ,4 ,CACHE_LINE_BURST ,32 ,0 }; +int keep_tab[] = {32 ,128 ,512 ,0 }; +int *get_bulk_ptr; +int *put_bulk_ptr; +int *keep_ptr;int ret;Same here, changes would be minimal.
app/test/test_mempool_perf.c launch_cores(struct rte_mempool *mp, unsigned · line 289
Quoted 1 line
app/test/test_mempool_perf.c +0
-0for (get_bulk_ptr = bulk_tab_get; *get_bulk_ptr; get_bulk_ptr++){
app/test/test_mempool_perf.c do_one_mempool_test(struct rte_mempool *mp,
Quoted 1 line
app/test/test_mempool_perf.c +0
-0@@ -266,13 +291,16 @@ do_one_mempool_test(struct rte_mempool *mp,unsigned int cores)
app/test/test_mempool_perf.c do_one_mempool_test(struct rte_mempool *mp, unsigned int cores)
Quoted 6 lines
app/test/test_mempool_perf.c +1
-2for (keep_ptr = keep_tab; *keep_ptr; keep_ptr++) { n_get_bulk = *get_bulk_ptr; - n_put_bulk = *put_bulk_ptr; - n_keep = *keep_ptr; + n_put_bulk = n_get_bulk > 0 ? *put_bulk_ptr :
n_get_bulk;app/test/test_mempool_perf.c do_one_mempool_test(struct rte_mempool *mp, · line 295
Quoted 1 line
app/test/test_mempool_perf.c +1
-0+ n_keep = n_get_bulk > 0 ? *keep_ptr : -
*keep_ptr;app/test/test_mempool_perf.c do_one_mempool_test(struct rte_mempool *mp, unsigned int cores) · line 299
Quoted 4 lines
app/test/test_mempool_perf.c +0
-0ret =launch_cores (mp, cores);if (ret <0 )return -1 ;No change would be required above (except
use_constant_values= 0). And below, we could simply add instead:/* replay test with constant values */ if (n_get_bulk == n_put_bulk) { use_constant_values = 1; ret = launch_cores(mp, cores); if (ret < 0) return -1; }
Quoted 2 lines
If you are ok with the proposed changes, I can directly submit a v2, since I already made them locally.
Please do! No need do the work one more time. :-)
Quoted 3 lines
Thanks,
Olivier
PATCH 1/1 mempool: test performance with constant n
In reply to Morten BrørupFrom: Morten Brørup <[email protected]>
"What gets measured gets done."
This patch adds mempool performance tests where the number of objects to put and get is constant at compile time, which may significantly improve the performance of these functions. [*]
Also, it is ensured that the array holding the object used for testing is cache line aligned, for maximum performance.
And finally, the following entries are added to the list of tests:
- Number of kept objects: 512
- Number of objects to get and to put: The number of pointers fitting into a cache line, i.e. 8 or 16
[*] Some example performance test (with cache) results:
get_bulk=4 put_bulk=4 keep=128 constant_n=false rate_persec=280480972 get_bulk=4 put_bulk=4 keep=128 constant_n=true rate_persec=622159462
get_bulk=8 put_bulk=8 keep=128 constant_n=false rate_persec=477967155 get_bulk=8 put_bulk=8 keep=128 constant_n=true rate_persec=917582643
get_bulk=32 put_bulk=32 keep=32 constant_n=false rate_persec=871248691 get_bulk=32 put_bulk=32 keep=32 constant_n=true rate_persec=1134021836
---
Hi Morten,
Here is the updated patch.
I launched the mempool_perf on my desktop machine, but I don't reproduce the numbers: constant or non-constant give almost the same rate on my machine (it's even worst with constants). I tested with your initial patch and with this one. Can you please try this patch, and/or give some details about your test environment? Here is what I get:
with your patch:mempool_autotest cache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=false rate_persec=152620236 mempool_autotest cache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=true rate_persec=144716595 mempool_autotest cache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=false rate_persec=306996838 mempool_autotest cache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=true rate_persec=287375359 mempool_autotest cache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=false rate_persec=977626723 mempool_autotest cache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=true rate_persec=963103944
with this patch:mempool_autotest cache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=0 rate_persec=156460646 mempool_autotest cache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=1 rate_persec=142173798 mempool_autotest cache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=0 rate_persec=312410111 mempool_autotest cache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=1 rate_persec=281699942 mempool_autotest cache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=0 rate_persec=983315247 mempool_autotest cache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=1 rate_persec=950350638
v2:
- use a flag instead of a negative value to enable tests with compile-time constant
- use a static inline function instead of a macro
- remove some "noise" (do not change variable type when not required)
Thanks,
Olivier
app/test/test_mempool_perf.c | 110 ++++++++++++++++++++++++----------- 1 file changed, 77 insertions(+), 33 deletions(-)
app/test/test_mempool_perf.c +77 -33
@@ -1,5 +1,6 @@/* SPDX-License-Identifier: BSD-3-Clause * Copyright(c) 2010-2014 Intel Corporation +* Copyright(c) 2022 SmartShare Systems */ #include <string.h>
@@ -55,19 +56,24 @@* * - Bulk get from 1 to 32 * - Bulk put from 1 to 32 +* - Bulk get and put from 1 to 32, compile time constant * * - Number of kept objects (*n_keep*) * * - 32 * - 128 +* - 512 */ #define N 65536#define TIME_S 5#define MEMPOOL_ELT_SIZE 2048 -#define MAX_KEEP 128 +#define MAX_KEEP 512#define MEMPOOL_SIZE ((rte_lcore_count()*(MAX_KEEP+RTE_MEMPOOL_CACHE_MAX_SIZE))-1) +/* Number of pointers fitting into one cache line. */ +#define CACHE_LINE_BURST (RTE_CACHE_LINE_SIZE / sizeof(uintptr_t)) +#define LOG_ERR () printf("test failed at %s():%d\n", __func__, __LINE__)#define RET_ERR () do { \ LOG_ERR(); \
@@ -91,6 +97,9 @@ static unsigned n_put_bulk;/* number of objects retrieved from mempool before putting them back */ static unsigned n_keep; +/* true if we want to test with constant n_get_bulk and n_put_bulk */ +static int use_constant_values; +/* number of enqueues / dequeues */ struct mempool_test_stats {uint64_t enq_count;
@@ -111,11 +120,43 @@ my_obj_init(struct rte_mempool *mp, __rte_unused void *arg, *objnum = i; } +static __rte_always_inline int +test_loop (struct rte_mempool *mp,struct rte_mempool_cache *cache, +unsigned int x_keep,unsigned int x_get_bulk,unsigned int x_put_bulk) +{ + void *obj_table[MAX_KEEP] __rte_cache_aligned; +unsigned int idx; +unsigned int i; +int ret; + +for (i =0 ;likely (i < (N / x_keep)); i++) { +/* get x_keep objects by bulk of x_get_bulk */ +for (idx =0 ; idx < x_keep; idx += x_get_bulk) { + ret =rte_mempool_generic_get (mp, + &obj_table[idx], + x_get_bulk, + cache); +if (unlikely (ret <0 )) { +rte_mempool_dump (stdout, mp); +return ret; + } + } + +/* put the objects back by bulk of x_put_bulk */ +for (idx =0 ; idx < x_keep; idx += x_put_bulk) { +rte_mempool_generic_put (mp, + &obj_table[idx], + x_put_bulk, + cache); + } + } + +return 0 ; +} +static int per_lcore_mempool_test (void *arg) { -void *obj_table[MAX_KEEP ]; -unsigned i, idx;struct rte_mempool *mp = arg;unsigned lcore_id =rte_lcore_id ();int ret =0 ;
@@ -139,6 +180,9 @@ per_lcore_mempool_test(void *arg)GOTO_ERR (ret, out);if (((n_keep / n_put_bulk) * n_put_bulk) != n_keep)GOTO_ERR (ret, out); +/* for constant n, n_get_bulk and n_put_bulk must be the same */ +if (use_constant_values && n_put_bulk != n_get_bulk) +GOTO_ERR (ret, out); stats[lcore_id].enq_count =0 ;
@@ -149,31 +193,23 @@ per_lcore_mempool_test(void *arg) start_cycles =rte_get_timer_cycles ();while (time_diff/hz <TIME_S ) { -for (i =0 ;likely (i < (N /n_keep)); i++) { -/* get n_keep objects by bulk of n_bulk */ - idx =0 ; -while (idx < n_keep) { - ret =rte_mempool_generic_get (mp, - &obj_table[idx], - n_get_bulk, - cache); -if (unlikely (ret <0 )) { -rte_mempool_dump (stdout, mp); -/* in this case, objects are lost... */ -GOTO_ERR (ret, out); - } - idx += n_get_bulk; - } +if (!use_constant_values) + ret =test_loop (mp, cache, n_keep, n_get_bulk, n_put_bulk); +else if (n_get_bulk ==1 ) + ret =test_loop (mp, cache, n_keep,1 ,1 ); +else if (n_get_bulk ==4 ) + ret =test_loop (mp, cache, n_keep,4 ,4 ); +else if (n_get_bulk ==CACHE_LINE_BURST ) + ret =test_loop (mp, cache, n_keep, +CACHE_LINE_BURST ,CACHE_LINE_BURST ); +else if (n_get_bulk ==32 ) + ret =test_loop (mp, cache, n_keep,32 ,32 ); +else + ret =-1 ; + +if (ret <0 ) +GOTO_ERR (ret, out); -/* put the objects back */ - idx =0 ; -while (idx < n_keep) { -rte_mempool_generic_put (mp, &obj_table[idx], - n_put_bulk, - cache); - idx += n_put_bulk; - } - } end_cycles =rte_get_timer_cycles (); time_diff = end_cycles - start_cycles; stats[lcore_id].enq_count +=N ;
@@ -203,10 +239,10 @@ launch_cores(struct rte_mempool *mp, unsigned int cores)memset (stats,0 ,sizeof (stats));printf ("mempool_autotest cache=%u cores=%u n_get_bulk=%u " -"n_put_bulk=%u n_keep=%u " , +"n_put_bulk=%u n_keep=%u constant_n=%u " , use_external_cache ? external_cache_size : (unsigned ) mp->cache_size, - cores, n_get_bulk, n_put_bulk, n_keep); + cores, n_get_bulk, n_put_bulk, n_keep, use_constant_values);if (rte_mempool_avail_count (mp) !=MEMPOOL_SIZE ) {printf ("mempool is not full\n" );
@@ -253,9 +289,9 @@ launch_cores(struct rte_mempool *mp, unsigned int cores)static int do_one_mempool_test (struct rte_mempool *mp,unsigned int cores) { -unsigned bulk_tab_get[] = {1 ,4 ,32 ,0 }; -unsigned bulk_tab_put[] = {1 ,4 ,32 ,0 }; -unsigned keep_tab[] = {32 ,128 ,0 }; +unsigned int bulk_tab_get[] = {1 ,4 ,CACHE_LINE_BURST ,32 ,0 }; +unsigned int bulk_tab_put[] = {1 ,4 ,CACHE_LINE_BURST ,32 ,0 }; +unsigned int keep_tab[] = {32 ,128 ,512 ,0 };unsigned *get_bulk_ptr;unsigned *put_bulk_ptr;unsigned *keep_ptr;
@@ -265,13 +301,21 @@ do_one_mempool_test(struct rte_mempool *mp, unsigned int cores)for (put_bulk_ptr = bulk_tab_put; *put_bulk_ptr; put_bulk_ptr++) {for (keep_ptr = keep_tab; *keep_ptr; keep_ptr++) { + use_constant_values =0 ; n_get_bulk = *get_bulk_ptr; n_put_bulk = *put_bulk_ptr; n_keep = *keep_ptr; ret =launch_cores (mp, cores); -if (ret <0 )return -1 ; + +/* replay test with constant values */ +if (n_get_bulk == n_put_bulk) { + use_constant_values =1 ; + ret =launch_cores (mp, cores); +if (ret <0 ) +return -1 ; + } } } }
--
2.30.2
Quoted 21 lines · Olivier Matz
From: Morten Brørup <[email protected]>
"What gets measured gets done."
This patch adds mempool performance tests where the number of objects to put and get is constant at compile time, which may significantly improve the performance of these functions. [*]
Also, it is ensured that the array holding the object used for testing is cache line aligned, for maximum performance.
And finally, the following entries are added to the list of tests:
- Number of kept objects: 512
- Number of objects to get and to put: The number of pointers fitting into a cache line, i.e. 8 or 16[*] Some example performance test (with cache) results:
get_bulk=4 put_bulk=4 keep=128 constant_n=false rate_persec=280480972 get_bulk=4 put_bulk=4 keep=128 constant_n=true rate_persec=622159462
Quoted 8 lines
get_bulk=8 put_bulk=8 keep=128 constant_n=false rate_persec=477967155 get_bulk=8 put_bulk=8 keep=128 constant_n=true rate_persec=917582643
get_bulk=32 put_bulk=32 keep=32 constant_n=false rate_persec=871248691 get_bulk=32 put_bulk=32 keep=32 constant_n=true rate_persec=1134021836
Signed-off-by Morten Brørup <[email protected]>Signed-off-by Olivier Matz <[email protected]>
Sorry, wrong title, "v2" is missing. Please ignore, I'm resending it.
PATCH v2 1/1 mempool: test performance with constant n
In reply to Morten BrørupFrom: Morten Brørup <[email protected]>
"What gets measured gets done."
This patch adds mempool performance tests where the number of objects to put and get is constant at compile time, which may significantly improve the performance of these functions. [*]
Also, it is ensured that the array holding the object used for testing is cache line aligned, for maximum performance.
And finally, the following entries are added to the list of tests:
- Number of kept objects: 512
- Number of objects to get and to put: The number of pointers fitting into a cache line, i.e. 8 or 16
[*] Some example performance test (with cache) results:
get_bulk=4 put_bulk=4 keep=128 constant_n=false rate_persec=280480972 get_bulk=4 put_bulk=4 keep=128 constant_n=true rate_persec=622159462
get_bulk=8 put_bulk=8 keep=128 constant_n=false rate_persec=477967155 get_bulk=8 put_bulk=8 keep=128 constant_n=true rate_persec=917582643
get_bulk=32 put_bulk=32 keep=32 constant_n=false rate_persec=871248691 get_bulk=32 put_bulk=32 keep=32 constant_n=true rate_persec=1134021836
---
Hi Morten,
Here is the updated patch.
I launched the mempool_perf on my desktop machine, but I don't reproduce the numbers: constant or non-constant give almost the same rate on my machine (it's even worst with constants). I tested with your initial patch and with this one. Can you please try this patch, and/or give some details about your test environment? Here is what I get:
with your patch:mempool_autotest cache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=false rate_persec=152620236 mempool_autotest cache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=true rate_persec=144716595 mempool_autotest cache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=false rate_persec=306996838 mempool_autotest cache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=true rate_persec=287375359 mempool_autotest cache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=false rate_persec=977626723 mempool_autotest cache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=true rate_persec=963103944
with this patch:mempool_autotest cache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=0 rate_persec=156460646 mempool_autotest cache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=1 rate_persec=142173798 mempool_autotest cache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=0 rate_persec=312410111 mempool_autotest cache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=1 rate_persec=281699942 mempool_autotest cache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=0 rate_persec=983315247 mempool_autotest cache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=1 rate_persec=950350638
v2:
- use a flag instead of a negative value to enable tests with compile-time constant
- use a static inline function instead of a macro
- remove some "noise" (do not change variable type when not required)
Thanks,
Olivier
app/test/test_mempool_perf.c | 110 ++++++++++++++++++++++++----------- 1 file changed, 77 insertions(+), 33 deletions(-)
app/test/test_mempool_perf.c +77 -33
@@ -1,5 +1,6 @@/* SPDX-License-Identifier: BSD-3-Clause * Copyright(c) 2010-2014 Intel Corporation +* Copyright(c) 2022 SmartShare Systems */ #include <string.h>
@@ -55,19 +56,24 @@* * - Bulk get from 1 to 32 * - Bulk put from 1 to 32 +* - Bulk get and put from 1 to 32, compile time constant * * - Number of kept objects (*n_keep*) * * - 32 * - 128 +* - 512 */ #define N 65536#define TIME_S 5#define MEMPOOL_ELT_SIZE 2048 -#define MAX_KEEP 128 +#define MAX_KEEP 512#define MEMPOOL_SIZE ((rte_lcore_count()*(MAX_KEEP+RTE_MEMPOOL_CACHE_MAX_SIZE))-1) +/* Number of pointers fitting into one cache line. */ +#define CACHE_LINE_BURST (RTE_CACHE_LINE_SIZE / sizeof(uintptr_t)) +#define LOG_ERR () printf("test failed at %s():%d\n", __func__, __LINE__)#define RET_ERR () do { \ LOG_ERR(); \
@@ -91,6 +97,9 @@ static unsigned n_put_bulk;/* number of objects retrieved from mempool before putting them back */ static unsigned n_keep; +/* true if we want to test with constant n_get_bulk and n_put_bulk */ +static int use_constant_values; +/* number of enqueues / dequeues */ struct mempool_test_stats {uint64_t enq_count;
@@ -111,11 +120,43 @@ my_obj_init(struct rte_mempool *mp, __rte_unused void *arg, *objnum = i; } +static __rte_always_inline int +test_loop (struct rte_mempool *mp,struct rte_mempool_cache *cache, +unsigned int x_keep,unsigned int x_get_bulk,unsigned int x_put_bulk) +{ + void *obj_table[MAX_KEEP] __rte_cache_aligned; +unsigned int idx; +unsigned int i; +int ret; + +for (i =0 ;likely (i < (N / x_keep)); i++) { +/* get x_keep objects by bulk of x_get_bulk */ +for (idx =0 ; idx < x_keep; idx += x_get_bulk) { + ret =rte_mempool_generic_get (mp, + &obj_table[idx], + x_get_bulk, + cache); +if (unlikely (ret <0 )) { +rte_mempool_dump (stdout, mp); +return ret; + } + } + +/* put the objects back by bulk of x_put_bulk */ +for (idx =0 ; idx < x_keep; idx += x_put_bulk) { +rte_mempool_generic_put (mp, + &obj_table[idx], + x_put_bulk, + cache); + } + } + +return 0 ; +} +static int per_lcore_mempool_test (void *arg) { -void *obj_table[MAX_KEEP ]; -unsigned i, idx;struct rte_mempool *mp = arg;unsigned lcore_id =rte_lcore_id ();int ret =0 ;
@@ -139,6 +180,9 @@ per_lcore_mempool_test(void *arg)GOTO_ERR (ret, out);if (((n_keep / n_put_bulk) * n_put_bulk) != n_keep)GOTO_ERR (ret, out); +/* for constant n, n_get_bulk and n_put_bulk must be the same */ +if (use_constant_values && n_put_bulk != n_get_bulk) +GOTO_ERR (ret, out); stats[lcore_id].enq_count =0 ;
@@ -149,31 +193,23 @@ per_lcore_mempool_test(void *arg) start_cycles =rte_get_timer_cycles ();while (time_diff/hz <TIME_S ) { -for (i =0 ;likely (i < (N /n_keep)); i++) { -/* get n_keep objects by bulk of n_bulk */ - idx =0 ; -while (idx < n_keep) { - ret =rte_mempool_generic_get (mp, - &obj_table[idx], - n_get_bulk, - cache); -if (unlikely (ret <0 )) { -rte_mempool_dump (stdout, mp); -/* in this case, objects are lost... */ -GOTO_ERR (ret, out); - } - idx += n_get_bulk; - } +if (!use_constant_values) + ret =test_loop (mp, cache, n_keep, n_get_bulk, n_put_bulk); +else if (n_get_bulk ==1 ) + ret =test_loop (mp, cache, n_keep,1 ,1 ); +else if (n_get_bulk ==4 ) + ret =test_loop (mp, cache, n_keep,4 ,4 ); +else if (n_get_bulk ==CACHE_LINE_BURST ) + ret =test_loop (mp, cache, n_keep, +CACHE_LINE_BURST ,CACHE_LINE_BURST ); +else if (n_get_bulk ==32 ) + ret =test_loop (mp, cache, n_keep,32 ,32 ); +else + ret =-1 ; + +if (ret <0 ) +GOTO_ERR (ret, out); -/* put the objects back */ - idx =0 ; -while (idx < n_keep) { -rte_mempool_generic_put (mp, &obj_table[idx], - n_put_bulk, - cache); - idx += n_put_bulk; - } - } end_cycles =rte_get_timer_cycles (); time_diff = end_cycles - start_cycles; stats[lcore_id].enq_count +=N ;
@@ -203,10 +239,10 @@ launch_cores(struct rte_mempool *mp, unsigned int cores)memset (stats,0 ,sizeof (stats));printf ("mempool_autotest cache=%u cores=%u n_get_bulk=%u " -"n_put_bulk=%u n_keep=%u " , +"n_put_bulk=%u n_keep=%u constant_n=%u " , use_external_cache ? external_cache_size : (unsigned ) mp->cache_size, - cores, n_get_bulk, n_put_bulk, n_keep); + cores, n_get_bulk, n_put_bulk, n_keep, use_constant_values);if (rte_mempool_avail_count (mp) !=MEMPOOL_SIZE ) {printf ("mempool is not full\n" );
@@ -253,9 +289,9 @@ launch_cores(struct rte_mempool *mp, unsigned int cores)static int do_one_mempool_test (struct rte_mempool *mp,unsigned int cores) { -unsigned bulk_tab_get[] = {1 ,4 ,32 ,0 }; -unsigned bulk_tab_put[] = {1 ,4 ,32 ,0 }; -unsigned keep_tab[] = {32 ,128 ,0 }; +unsigned int bulk_tab_get[] = {1 ,4 ,CACHE_LINE_BURST ,32 ,0 }; +unsigned int bulk_tab_put[] = {1 ,4 ,CACHE_LINE_BURST ,32 ,0 }; +unsigned int keep_tab[] = {32 ,128 ,512 ,0 };unsigned *get_bulk_ptr;unsigned *put_bulk_ptr;unsigned *keep_ptr;
@@ -265,13 +301,21 @@ do_one_mempool_test(struct rte_mempool *mp, unsigned int cores)for (put_bulk_ptr = bulk_tab_put; *put_bulk_ptr; put_bulk_ptr++) {for (keep_ptr = keep_tab; *keep_ptr; keep_ptr++) { + use_constant_values =0 ; n_get_bulk = *get_bulk_ptr; n_put_bulk = *put_bulk_ptr; n_keep = *keep_ptr; ret =launch_cores (mp, cores); -if (ret <0 )return -1 ; + +/* replay test with constant values */ +if (n_get_bulk == n_put_bulk) { + use_constant_values =1 ; + ret =launch_cores (mp, cores); +if (ret <0 ) +return -1 ; + } } } }
--
2.30.2
Quoted 40 lines
From: Olivier Matz [mailto:[email protected]]
Sent: Monday, 24 January 2022 16.00From: Morten Brørup <[email protected]>
"What gets measured gets done."
This patch adds mempool performance tests where the number of objects to
put and get is constant at compile time, which may significantly improve
the performance of these functions. [*]Also, it is ensured that the array holding the object used for testing is cache line aligned, for maximum performance.
And finally, the following entries are added to the list of tests:
- Number of kept objects: 512
- Number of objects to get and to put: The number of pointers fitting into a cache line, i.e. 8 or 16[*] Some example performance test (with cache) results:
get_bulk=4 put_bulk=4 keep=128 constant_n=false rate_persec=280480972 get_bulk=4 put_bulk=4 keep=128 constant_n=true rate_persec=622159462
get_bulk=8 put_bulk=8 keep=128 constant_n=false rate_persec=477967155 get_bulk=8 put_bulk=8 keep=128 constant_n=true rate_persec=917582643
get_bulk=32 put_bulk=32 keep=32 constant_n=false rate_persec=871248691 get_bulk=32 put_bulk=32 keep=32 constant_n=true rate_persec=1134021836
Signed-off-by Morten Brørup <[email protected]>Signed-off-by Olivier Matz <[email protected]>---
Hi Morten,
Here is the updated patch.
Quoted 7 lines
I launched the
mempool_perfon my desktop machine, but I don't reproduce the numbers: constant or
non-constant give almost the same rate on my machine (it's even worst with constants). I tested with
your initial patch and with this one. Can you please try this patch, and/or give some details about
your test environment?
Test environment:
VMware virtual machine running Ubuntu 20.04.3 LTS.
4 CPUs and 8 GB RAM assigned.
The physical CPU is a Xeon E5-2620 v4 with plenty of RAM.
Although other VMs are running on the same server, it is not very oversubscribed.
Hugepages established with:usertools/dpdk-hugepages.py -p 2M --setup 2G
Build steps:
meson -Dplatform=generic work
cd work
ninja
Quoted 15 lines
Here is what I get:
with your patch:
mempool_autotestcache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=false rate_persec=152620236mempool_autotestcache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=true rate_persec=144716595mempool_autotestcache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=false rate_persec=306996838mempool_autotestcache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=true rate_persec=287375359mempool_autotestcache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=false rate_persec=977626723mempool_autotestcache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=true rate_persec=963103944
My test results were with an experimental, optimized version of the mempool library, which showed a larger difference. (This was the reason for updating the perf test - to measure the effects of optimizing the mempool library.)
However, testing the patch (version 1) with a brand new git checkout still shows a huge difference, e.g.:
mempool_autotest cache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=false rate_persec=501009612 mempool_autotest cache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=true rate_persec=799014912
You should also see a significant difference when testing.
My rate_persec without constant n is 3 x yours (501 M vs. 156 M ops/s), so the baseline seems wrong! I don't think our server rig is so much faster than your desktop machine. Perhaps mempool debug, telemetry or other background noise is polluting your test.
Quoted 244 lines
with this patch:
mempool_autotestcache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=0 rate_persec=156460646mempool_autotestcache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=1 rate_persec=142173798mempool_autotestcache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=0 rate_persec=312410111mempool_autotestcache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=1 rate_persec=281699942mempool_autotestcache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=0 rate_persec=983315247mempool_autotestcache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=1 rate_persec=950350638v2:
- use a flag instead of a negative value to enable tests with compile-time constant
- use a static inline function instead of a macro
- remove some "noise" (do not change variable type when not required)Thanks,
Olivierapp/test/test_mempool_perf.c | 110 ++++++++++++++++++++++++----------- 1 file changed, 77 insertions(+), 33 deletions(-)app/test/test_mempool_perf.c +7
-1@@ -1,5 +1,6 @@/* SPDX-License-Identifier: BSD-3-Clause * Copyright(c) 2010-2014 Intel Corporation +* Copyright(c) 2022 SmartShare Systems */ #include <string.h> @@ -55,19 +56,24 @@* * - Bulk get from 1 to 32 * - Bulk put from 1 to 32 +* - Bulk get and put from 1 to 32, compile time constant * * - Number of kept objects (*n_keep*) * * - 32 * - 128 +* - 512 */ #define N 65536#define TIME_S 5#define MEMPOOL_ELT_SIZE 2048 -#define MAX_KEEP 128 +#define MAX_KEEP 512#define MEMPOOL_SIZE ((rte_lcore_count ()*(MAX_KEEP +RTE_MEMPOOL_CACHE_MAX_SIZE ))-1 ) +/* Number of pointers fitting into one cache line. */ +#define CACHE_LINE_BURST (RTE_CACHE_LINE_SIZE / sizeof(uintptr_t)) +#define LOG_ERR () printf("test failed at %s():%d\n", __func__, __LINE__)#define
RET_ERR()do { \app/test/test_mempool_perf.c +3
-0LOG_ERR (); \@@ -91,6 +97,9 @@ static unsigned n_put_bulk;/* number of objects retrieved from mempool before putting them back */ static unsigned n_keep; +/* true if we want to test with constant n_get_bulk and n_put_bulk */ +static int use_constant_values; +/* number of enqueues / dequeues */ struct mempool_test_stats {app/test/test_mempool_perf.c +34
-2uint64_t enq_count;@@ -111,11 +120,43 @@ my_obj_init(struct rte_mempool *mp, __rte_unusedvoid *arg, *objnum = i; } +static __rte_always_inline int +test_loop (struct rte_mempool *mp,struct rte_mempool_cache *cache, +unsigned int x_keep,unsigned int x_get_bulk,unsigned int x_put_bulk) +{ + void *obj_table[MAX_KEEP] __rte_cache_aligned; +unsigned int idx; +unsigned int i; +int ret; + +for (i =0 ;likely (i < (N / x_keep)); i++) { +/* get x_keep objects by bulk of x_get_bulk */ +for (idx =0 ; idx < x_keep; idx += x_get_bulk) { + ret =rte_mempool_generic_get (mp, + &obj_table[idx], + x_get_bulk, + cache); +if (unlikely (ret <0 )) { +rte_mempool_dump (stdout, mp); +return ret; + } + } + +/* put the objects back by bulk of x_put_bulk */ +for (idx =0 ; idx < x_keep; idx += x_put_bulk) { +rte_mempool_generic_put (mp, + &obj_table[idx], + x_put_bulk, + cache); + } + } + +return 0 ; +} +static int per_lcore_mempool_test (void *arg) { -void *obj_table[MAX_KEEP ]; -unsigned i, idx;struct rte_mempool *mp = arg;app/test/test_mempool_perf.c +19
-24unsigned lcore_id =rte_lcore_id ();int ret =0 ;@@ -139,6 +180,9 @@ per_lcore_mempool_test(void *arg)GOTO_ERR (ret, out);if (((n_keep / n_put_bulk) * n_put_bulk) != n_keep)GOTO_ERR (ret, out); +/* for constant n, n_get_bulk and n_put_bulk must be the same */ +if (use_constant_values && n_put_bulk != n_get_bulk) +GOTO_ERR (ret, out); stats[lcore_id].enq_count =0 ;@@ -149,31 +193,23 @@ per_lcore_mempool_test(void *arg) start_cycles =rte_get_timer_cycles ();while (time_diff/hz <TIME_S ) { -for (i =0 ;likely (i < (N /n_keep)); i++) { -/* get n_keep objects by bulk of n_bulk */ - idx =0 ; -while (idx < n_keep) { - ret =rte_mempool_generic_get (mp, - &obj_table[idx], - n_get_bulk, - cache); -if (unlikely (ret <0 )) { -rte_mempool_dump (stdout, mp); -/* in this case, objects are lost... */ -GOTO_ERR (ret, out); - } - idx += n_get_bulk; - } +if (!use_constant_values) + ret =test_loop (mp, cache, n_keep, n_get_bulk, n_put_bulk); +else if (n_get_bulk ==1 ) + ret =test_loop (mp, cache, n_keep,1 ,1 ); +else if (n_get_bulk ==4 ) + ret =test_loop (mp, cache, n_keep,4 ,4 ); +else if (n_get_bulk ==CACHE_LINE_BURST ) + ret =test_loop (mp, cache, n_keep, +CACHE_LINE_BURST ,CACHE_LINE_BURST ); +else if (n_get_bulk ==32 ) + ret =test_loop (mp, cache, n_keep,32 ,32 ); +else + ret =-1 ; + +if (ret <0 ) +GOTO_ERR (ret, out); -/* put the objects back */ - idx =0 ; -while (idx < n_keep) { -rte_mempool_generic_put (mp, &obj_table[idx], - n_put_bulk, - cache); - idx += n_put_bulk; - } - } end_cycles =rte_get_timer_cycles (); time_diff = end_cycles - start_cycles;app/test/test_mempool_perf.c +2
-2stats[lcore_id].enq_count +=N ;@@ -203,10 +239,10 @@ launch_cores(struct rte_mempool *mp, unsigned int cores) memset(stats, 0, sizeof(stats));printf ("mempool_autotest cache=%u cores=%u n_get_bulk=%u " -"n_put_bulk=%u n_keep=%u " , +"n_put_bulk=%u n_keep=%u constant_n=%u " , use_external_cache ? external_cache_size : (unsigned ) mp->cache_size, - cores, n_get_bulk, n_put_bulk, n_keep); + cores, n_get_bulk, n_put_bulk, n_keep, use_constant_values);app/test/test_mempool_perf.c +3
-3if (rte_mempool_avail_count (mp) !=MEMPOOL_SIZE ) {printf ("mempool is not full\n" );@@ -253,9 +289,9 @@ launch_cores(struct rte_mempool *mp, unsigned int cores) static intdo_one_mempool_test (struct rte_mempool *mp,unsigned int cores) { -unsigned bulk_tab_get[] = {1 ,4 ,32 ,0 }; -unsigned bulk_tab_put[] = {1 ,4 ,32 ,0 }; -unsigned keep_tab[] = {32 ,128 ,0 }; +unsigned int bulk_tab_get[] = {1 ,4 ,CACHE_LINE_BURST ,32 ,0 }; +unsigned int bulk_tab_put[] = {1 ,4 ,CACHE_LINE_BURST ,32 ,0 }; +unsigned int keep_tab[] = {32 ,128 ,512 ,0 };unsigned *get_bulk_ptr;unsigned *put_bulk_ptr;app/test/test_mempool_perf.c +9
-1unsigned *keep_ptr;@@ -265,13 +301,21 @@ do_one_mempool_test(struct rte_mempool *mp, unsigned int cores)for (put_bulk_ptr = bulk_tab_put; *put_bulk_ptr; put_bulk_ptr++) {for (keep_ptr = keep_tab; *keep_ptr; keep_ptr++) { + use_constant_values =0 ; n_get_bulk = *get_bulk_ptr; n_put_bulk = *put_bulk_ptr; n_keep = *keep_ptr; ret =launch_cores (mp, cores); -if (ret <0 )return -1 ; + +/* replay test with constant values */ +if (n_get_bulk == n_put_bulk) { + use_constant_values =1 ; + ret =launch_cores (mp, cores); +if (ret <0 ) +return -1 ; + } }app/test/test_mempool_perf.c +0
-1} } --2.30.2
Hi Morten,
Quoted 80 lines · Morten Brørup
Quoted 40 lines
From: Olivier Matz [mailto:[email protected]]
Sent: Monday, 24 January 2022 16.00From: Morten Brørup <[email protected]>
"What gets measured gets done."
This patch adds mempool performance tests where the number of objects to
put and get is constant at compile time, which may significantly improve
the performance of these functions. [*]Also, it is ensured that the array holding the object used for testing is cache line aligned, for maximum performance.
And finally, the following entries are added to the list of tests:
- Number of kept objects: 512
- Number of objects to get and to put: The number of pointers fitting into a cache line, i.e. 8 or 16[*] Some example performance test (with cache) results:
get_bulk=4 put_bulk=4 keep=128 constant_n=false rate_persec=280480972 get_bulk=4 put_bulk=4 keep=128 constant_n=true rate_persec=622159462
get_bulk=8 put_bulk=8 keep=128 constant_n=false rate_persec=477967155 get_bulk=8 put_bulk=8 keep=128 constant_n=true rate_persec=917582643
get_bulk=32 put_bulk=32 keep=32 constant_n=false rate_persec=871248691 get_bulk=32 put_bulk=32 keep=32 constant_n=true rate_persec=1134021836
Signed-off-by Morten Brørup <[email protected]>Signed-off-by Olivier Matz <[email protected]>---
Hi Morten,
Here is the updated patch.
Quoted 7 lines
I launched the
mempool_perfon my desktop machine, but I don't reproduce the numbers: constant or
non-constant give almost the same rate on my machine (it's even worst with constants). I tested with
your initial patch and with this one. Can you please try this patch, and/or give some details about
your test environment?Test environment:
VMware virtual machine running Ubuntu 20.04.3 LTS.
4 CPUs and 8 GB RAM assigned.
The physical CPU is a Xeon E5-2620 v4 with plenty of RAM.
Although other VMs are running on the same server, it is not very oversubscribed.Hugepages established with:
usertools/dpdk-hugepages.py-p 2M --setup 2GBuild steps:
meson -Dplatform=generic work
cd work
ninjaQuoted 15 lines
Here is what I get:
with your patch:
mempool_autotestcache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=false rate_persec=152620236mempool_autotestcache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=true rate_persec=144716595mempool_autotestcache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=false rate_persec=306996838mempool_autotestcache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=true rate_persec=287375359mempool_autotestcache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=false rate_persec=977626723mempool_autotestcache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=true rate_persec=963103944My test results were with an experimental, optimized version of the mempool library, which showed a larger difference. (This was the reason for updating the perf test - to measure the effects of optimizing the mempool library.)
Quoted 8 lines
However, testing the patch (version 1) with a brand new git checkout still shows a huge difference, e.g.:
mempool_autotestcache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=false rate_persec=501009612mempool_autotestcache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=true rate_persec=799014912You should also see a significant difference when testing.
My
rate_persecwithout constant n is 3 x yours (501 M vs. 156 M ops/s), so the baseline seems wrong! I don't think our server rig is so much faster than your desktop machine. Perhaps mempool debug, telemetry or other background noise is polluting your test.
Sorry, I just realized that I was indeed using a "debugoptimzed" build. It's much better in release mode.
mempool_autotest cache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=0 rate_persec=1425473536 mempool_autotest cache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=1 rate_persec=2159660236 mempool_autotest cache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=0 rate_persec=2796342476 mempool_autotest cache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=1 rate_persec=4351577292 mempool_autotest cache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=0 rate_persec=8589777300 mempool_autotest cache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=1 rate_persec=13560971258
Thanks,
Olivier
Quoted 246 lines
Quoted 245 lines
with this patch:
mempool_autotestcache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=0 rate_persec=156460646mempool_autotestcache=512 cores=1 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=1 rate_persec=142173798mempool_autotestcache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=0 rate_persec=312410111mempool_autotestcache=512 cores=2 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=1 rate_persec=281699942mempool_autotestcache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=0 rate_persec=983315247mempool_autotestcache=512 cores=12 n_get_bulk=8 n_put_bulk=8 n_keep=128 constant_n=1 rate_persec=950350638v2:
- use a flag instead of a negative value to enable tests with compile-time constant
- use a static inline function instead of a macro
- remove some "noise" (do not change variable type when not required)Thanks,
Olivierapp/test/test_mempool_perf.c | 110 ++++++++++++++++++++++++----------- 1 file changed, 77 insertions(+), 33 deletions(-)app/test/test_mempool_perf.c +7
-1@@ -1,5 +1,6 @@/* SPDX-License-Identifier: BSD-3-Clause * Copyright(c) 2010-2014 Intel Corporation +* Copyright(c) 2022 SmartShare Systems */ #include <string.h> @@ -55,19 +56,24 @@* * - Bulk get from 1 to 32 * - Bulk put from 1 to 32 +* - Bulk get and put from 1 to 32, compile time constant * * - Number of kept objects (*n_keep*) * * - 32 * - 128 +* - 512 */ #define N 65536#define TIME_S 5#define MEMPOOL_ELT_SIZE 2048 -#define MAX_KEEP 128 +#define MAX_KEEP 512#define MEMPOOL_SIZE ((rte_lcore_count ()*(MAX_KEEP +RTE_MEMPOOL_CACHE_MAX_SIZE ))-1 ) +/* Number of pointers fitting into one cache line. */ +#define CACHE_LINE_BURST (RTE_CACHE_LINE_SIZE / sizeof(uintptr_t)) +#define LOG_ERR () printf("test failed at %s():%d\n", __func__, __LINE__)#define
RET_ERR()do { \app/test/test_mempool_perf.c +3
-0LOG_ERR (); \@@ -91,6 +97,9 @@ static unsigned n_put_bulk;/* number of objects retrieved from mempool before putting them back */ static unsigned n_keep; +/* true if we want to test with constant n_get_bulk and n_put_bulk */ +static int use_constant_values; +/* number of enqueues / dequeues */ struct mempool_test_stats {app/test/test_mempool_perf.c +34
-2uint64_t enq_count;@@ -111,11 +120,43 @@ my_obj_init(struct rte_mempool *mp, __rte_unusedvoid *arg, *objnum = i; } +static __rte_always_inline int +test_loop (struct rte_mempool *mp,struct rte_mempool_cache *cache, +unsigned int x_keep,unsigned int x_get_bulk,unsigned int x_put_bulk) +{ + void *obj_table[MAX_KEEP] __rte_cache_aligned; +unsigned int idx; +unsigned int i; +int ret; + +for (i =0 ;likely (i < (N / x_keep)); i++) { +/* get x_keep objects by bulk of x_get_bulk */ +for (idx =0 ; idx < x_keep; idx += x_get_bulk) { + ret =rte_mempool_generic_get (mp, + &obj_table[idx], + x_get_bulk, + cache); +if (unlikely (ret <0 )) { +rte_mempool_dump (stdout, mp); +return ret; + } + } + +/* put the objects back by bulk of x_put_bulk */ +for (idx =0 ; idx < x_keep; idx += x_put_bulk) { +rte_mempool_generic_put (mp, + &obj_table[idx], + x_put_bulk, + cache); + } + } + +return 0 ; +} +static int per_lcore_mempool_test (void *arg) { -void *obj_table[MAX_KEEP ]; -unsigned i, idx;struct rte_mempool *mp = arg;app/test/test_mempool_perf.c +19
-24unsigned lcore_id =rte_lcore_id ();int ret =0 ;@@ -139,6 +180,9 @@ per_lcore_mempool_test(void *arg)GOTO_ERR (ret, out);if (((n_keep / n_put_bulk) * n_put_bulk) != n_keep)GOTO_ERR (ret, out); +/* for constant n, n_get_bulk and n_put_bulk must be the same */ +if (use_constant_values && n_put_bulk != n_get_bulk) +GOTO_ERR (ret, out); stats[lcore_id].enq_count =0 ;@@ -149,31 +193,23 @@ per_lcore_mempool_test(void *arg) start_cycles =rte_get_timer_cycles ();while (time_diff/hz <TIME_S ) { -for (i =0 ;likely (i < (N /n_keep)); i++) { -/* get n_keep objects by bulk of n_bulk */ - idx =0 ; -while (idx < n_keep) { - ret =rte_mempool_generic_get (mp, - &obj_table[idx], - n_get_bulk, - cache); -if (unlikely (ret <0 )) { -rte_mempool_dump (stdout, mp); -/* in this case, objects are lost... */ -GOTO_ERR (ret, out); - } - idx += n_get_bulk; - } +if (!use_constant_values) + ret =test_loop (mp, cache, n_keep, n_get_bulk, n_put_bulk); +else if (n_get_bulk ==1 ) + ret =test_loop (mp, cache, n_keep,1 ,1 ); +else if (n_get_bulk ==4 ) + ret =test_loop (mp, cache, n_keep,4 ,4 ); +else if (n_get_bulk ==CACHE_LINE_BURST ) + ret =test_loop (mp, cache, n_keep, +CACHE_LINE_BURST ,CACHE_LINE_BURST ); +else if (n_get_bulk ==32 ) + ret =test_loop (mp, cache, n_keep,32 ,32 ); +else + ret =-1 ; + +if (ret <0 ) +GOTO_ERR (ret, out); -/* put the objects back */ - idx =0 ; -while (idx < n_keep) { -rte_mempool_generic_put (mp, &obj_table[idx], - n_put_bulk, - cache); - idx += n_put_bulk; - } - } end_cycles =rte_get_timer_cycles (); time_diff = end_cycles - start_cycles;app/test/test_mempool_perf.c +2
-2stats[lcore_id].enq_count +=N ;@@ -203,10 +239,10 @@ launch_cores(struct rte_mempool *mp, unsigned int cores) memset(stats, 0, sizeof(stats));printf ("mempool_autotest cache=%u cores=%u n_get_bulk=%u " -"n_put_bulk=%u n_keep=%u " , +"n_put_bulk=%u n_keep=%u constant_n=%u " , use_external_cache ? external_cache_size : (unsigned ) mp->cache_size, - cores, n_get_bulk, n_put_bulk, n_keep); + cores, n_get_bulk, n_put_bulk, n_keep, use_constant_values);app/test/test_mempool_perf.c +3
-3if (rte_mempool_avail_count (mp) !=MEMPOOL_SIZE ) {printf ("mempool is not full\n" );@@ -253,9 +289,9 @@ launch_cores(struct rte_mempool *mp, unsigned int cores) static intdo_one_mempool_test (struct rte_mempool *mp,unsigned int cores) { -unsigned bulk_tab_get[] = {1 ,4 ,32 ,0 }; -unsigned bulk_tab_put[] = {1 ,4 ,32 ,0 }; -unsigned keep_tab[] = {32 ,128 ,0 }; +unsigned int bulk_tab_get[] = {1 ,4 ,CACHE_LINE_BURST ,32 ,0 }; +unsigned int bulk_tab_put[] = {1 ,4 ,CACHE_LINE_BURST ,32 ,0 }; +unsigned int keep_tab[] = {32 ,128 ,512 ,0 };unsigned *get_bulk_ptr;unsigned *put_bulk_ptr;app/test/test_mempool_perf.c +9
-1unsigned *keep_ptr;@@ -265,13 +301,21 @@ do_one_mempool_test(struct rte_mempool *mp, unsigned int cores)for (put_bulk_ptr = bulk_tab_put; *put_bulk_ptr; put_bulk_ptr++) {for (keep_ptr = keep_tab; *keep_ptr; keep_ptr++) { + use_constant_values =0 ; n_get_bulk = *get_bulk_ptr; n_put_bulk = *put_bulk_ptr; n_keep = *keep_ptr; ret =launch_cores (mp, cores); -if (ret <0 )return -1 ; + +/* replay test with constant values */ +if (n_get_bulk == n_put_bulk) { + use_constant_values =1 ; + ret =launch_cores (mp, cores); +if (ret <0 ) +return -1 ; + } }app/test/test_mempool_perf.c +0
-1} } --2.30.2
Thomas Monjalon
Applied24/01/2022 15:59, Olivier Matz:
Quoted 21 lines
From: Morten Brørup <[email protected]>
"What gets measured gets done."
This patch adds mempool performance tests where the number of objects to put and get is constant at compile time, which may significantly improve the performance of these functions. [*]
Also, it is ensured that the array holding the object used for testing is cache line aligned, for maximum performance.
And finally, the following entries are added to the list of tests:
- Number of kept objects: 512
- Number of objects to get and to put: The number of pointers fitting into a cache line, i.e. 8 or 16[*] Some example performance test (with cache) results:
get_bulk=4 put_bulk=4 keep=128 constant_n=false rate_persec=280480972 get_bulk=4 put_bulk=4 keep=128 constant_n=true rate_persec=622159462
Quoted 8 lines
get_bulk=8 put_bulk=8 keep=128 constant_n=false rate_persec=477967155 get_bulk=8 put_bulk=8 keep=128 constant_n=true rate_persec=917582643
get_bulk=32 put_bulk=32 keep=32 constant_n=false rate_persec=871248691 get_bulk=32 put_bulk=32 keep=32 constant_n=true rate_persec=1134021836
Signed-off-by Morten Brørup <[email protected]>Signed-off-by Olivier Matz <[email protected]>
Applied, thanks.