From mboxrd@z Thu Jan 1 00:00:00 1970
Return-Path: <dev-bounces@dpdk.org>
Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124])
by inbox.dpdk.org (Postfix) with ESMTP id AD336464C7;
Mon, 31 Mar 2025 12:04:03 +0200 (CEST)
Received: from mails.dpdk.org (localhost [127.0.0.1])
by mails.dpdk.org (Postfix) with ESMTP id 3B9ED4066C;
Mon, 31 Mar 2025 12:03:55 +0200 (CEST)
Received: from dkmailrelay1.smartsharesystems.com
(smartserver.smartsharesystems.com [77.243.40.215])
by mails.dpdk.org (Postfix) with ESMTP id D597C40156
for <dev@dpdk.org>; Mon, 31 Mar 2025 12:03:52 +0200 (CEST)
Received: from smartserver.smartsharesystems.com
(smartserver.smartsharesys.local [192.168.4.10])
by dkmailrelay1.smartsharesystems.com (Postfix) with ESMTP id 9E187228C7;
Mon, 31 Mar 2025 12:03:52 +0200 (CEST)
Received: from dkrd4.smartsharesys.local ([192.168.4.26]) by
smartserver.smartsharesystems.com with Microsoft SMTPSVC(6.0.3790.4675);
Mon, 31 Mar 2025 12:03:52 +0200
From: =?UTF-8?q?Morten=20Br=C3=B8rup?= <mb@smartsharesystems.com>
To: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>,
Bruce Richardson <bruce.richardson@intel.com>, dev@dpdk.org
Cc: =?UTF-8?q?Morten=20Br=C3=B8rup?= <mb@smartsharesystems.com>
Subject: [PATCH v2 4/4] mempool perf test: test random bulk sizes
Date: Mon, 31 Mar 2025 10:03:42 +0000
Message-ID: <20250331100343.213430-5-mb@smartsharesystems.com>
X-Mailer: git-send-email 2.43.0
In-Reply-To: <20250331100343.213430-1-mb@smartsharesystems.com>
References: <20250228164858.274204-1-mb@smartsharesystems.com>
<20250331100343.213430-1-mb@smartsharesystems.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
X-OriginalArrivalTime: 31 Mar 2025 10:03:52.0359 (UTC)
FILETIME=[34A93B70:01DBA224]
X-BeenThere: dev@dpdk.org
X-Mailman-Version: 2.1.29
Precedence: list
List-Id: DPDK patches and discussions <dev.dpdk.org>
List-Unsubscribe: <https://mails.dpdk.org/options/dev>,
<mailto:dev-request@dpdk.org?subject=unsubscribe>
List-Archive: <http://mails.dpdk.org/archives/dev/>
List-Post: <mailto:dev@dpdk.org>
List-Help: <mailto:dev-request@dpdk.org?subject=help>
List-Subscribe: <https://mails.dpdk.org/listinfo/dev>,
<mailto:dev-request@dpdk.org?subject=subscribe>
Errors-To: dev-bounces@dpdk.org
Bulk requests to get or put objects in a mempool often vary in size.
A series of tests with pseudo random request sizes, to mitigate the
benefits of the CPU's dynamic branch predictor, was added.
Signed-off-by: Morten Brørup <mb@smartsharesystems.com>
Acked-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>
---
app/test/test_mempool_perf.c | 102 ++++++++++++++++++++++++++++++++---
1 file changed, 95 insertions(+), 7 deletions(-)
diff --git a/app/test/test_mempool_perf.c b/app/test/test_mempool_perf.c
index c252f1968a..40ecaa5049 100644
--- a/app/test/test_mempool_perf.c
+++ b/app/test/test_mempool_perf.c
@@ -33,6 +33,13 @@
* Mempool performance
* =======
*
+ * Each core get *n_keep* objects per bulk of a pseudorandom number
+ * between 1 and *n_max_bulk*.
+ * Objects are put back in the pool per bulk of a similar pseudorandom number.
+ * Note: The very low entropy of the randomization algorithm is harmless, because
+ * the sole purpose of randomization is to prevent the CPU's dynamic branch
+ * predictor from enhancing the test results.
+ *
* Each core get *n_keep* objects per bulk of *n_get_bulk*. Then,
* objects are put back in the pool per bulk of *n_put_bulk*.
*
@@ -52,7 +59,12 @@
* - Two cores with user-owned cache
* - Max. cores with user-owned cache
*
- * - Bulk size (*n_get_bulk*, *n_put_bulk*)
+ * - Pseudorandom max bulk size (*n_max_bulk*)
+ *
+ * - Max bulk from CACHE_LINE_BURST to 256, and RTE_MEMPOOL_CACHE_MAX_SIZE,
+ * where CACHE_LINE_BURST is the number of pointers fitting into one CPU cache line.
+ *
+ * - Fixed bulk size (*n_get_bulk*, *n_put_bulk*)
*
* - Bulk get from 1 to 256, and RTE_MEMPOOL_CACHE_MAX_SIZE
* - Bulk put from 1 to 256, and RTE_MEMPOOL_CACHE_MAX_SIZE
@@ -93,6 +105,9 @@ static unsigned int external_cache_size = RTE_MEMPOOL_CACHE_MAX_SIZE;
static RTE_ATOMIC(uint32_t) synchro;
+/* max random number of objects in one bulk operation (get and put) */
+static unsigned int n_max_bulk;
+
/* number of objects in one bulk operation (get or put) */
static unsigned int n_get_bulk;
static unsigned int n_put_bulk;
@@ -159,6 +174,50 @@ test_loop(struct rte_mempool *mp, struct rte_mempool_cache *cache,
return 0;
}
+static __rte_always_inline int
+test_loop_random(struct rte_mempool *mp, struct rte_mempool_cache *cache,
+ unsigned int x_keep, unsigned int x_max_bulk)
+{
+ alignas(RTE_CACHE_LINE_SIZE) void *obj_table[MAX_KEEP];
+ unsigned int idx;
+ unsigned int i;
+ unsigned int r = 0;
+ unsigned int x_bulk;
+ int ret;
+
+ for (i = 0; likely(i < (N / x_keep)); i++) {
+ /* get x_keep objects by bulk of random [1 .. x_max_bulk] */
+ for (idx = 0; idx < x_keep; idx += x_bulk, r++) {
+ /* Generate a pseudorandom number [1 .. x_max_bulk]. */
+ x_bulk = ((r ^ (r >> 2) ^ (r << 3)) & (x_max_bulk - 1)) + 1;
+ if (unlikely(idx + x_bulk > x_keep))
+ x_bulk = x_keep - idx;
+ ret = rte_mempool_generic_get(mp,
+ &obj_table[idx],
+ x_bulk,
+ cache);
+ if (unlikely(ret < 0)) {
+ rte_mempool_dump(stdout, mp);
+ return ret;
+ }
+ }
+
+ /* put the objects back by bulk of random [1 .. x_max_bulk] */
+ for (idx = 0; idx < x_keep; idx += x_bulk, r++) {
+ /* Generate a pseudorandom number [1 .. x_max_bulk]. */
+ x_bulk = ((r ^ (r >> 2) ^ (r << 3)) & (x_max_bulk - 1)) + 1;
+ if (unlikely(idx + x_bulk > x_keep))
+ x_bulk = x_keep - idx;
+ rte_mempool_generic_put(mp,
+ &obj_table[idx],
+ x_bulk,
+ cache);
+ }
+ }
+
+ return 0;
+}
+
static int
per_lcore_mempool_test(void *arg)
{
@@ -181,9 +240,9 @@ per_lcore_mempool_test(void *arg)
}
/* n_get_bulk and n_put_bulk must be divisors of n_keep */
- if (((n_keep / n_get_bulk) * n_get_bulk) != n_keep)
+ if (!n_max_bulk && (((n_keep / n_get_bulk) * n_get_bulk) != n_keep))
GOTO_ERR(ret, out);
- if (((n_keep / n_put_bulk) * n_put_bulk) != n_keep)
+ if (!n_max_bulk && (((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)
@@ -200,7 +259,9 @@ per_lcore_mempool_test(void *arg)
start_cycles = rte_get_timer_cycles();
while (time_diff/hz < TIME_S) {
- if (!use_constant_values)
+ if (n_max_bulk)
+ ret = test_loop_random(mp, cache, n_keep, n_max_bulk);
+ else 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);
@@ -261,9 +322,13 @@ launch_cores(struct rte_mempool *mp, unsigned int cores)
use_external_cache ? external_cache_size : (unsigned int) mp->cache_size,
cores,
n_keep);
- printf("n_get_bulk=%3u n_put_bulk=%3u constant_n=%u ",
- n_get_bulk, n_put_bulk,
- use_constant_values);
+ if (n_max_bulk)
+ printf("n_max_bulk=%3u ",
+ n_max_bulk);
+ else
+ printf("n_get_bulk=%3u n_put_bulk=%3u constant_n=%u ",
+ n_get_bulk, n_put_bulk,
+ use_constant_values);
if (rte_mempool_avail_count(mp) != MEMPOOL_SIZE) {
printf("mempool is not full\n");
@@ -312,16 +377,37 @@ launch_cores(struct rte_mempool *mp, unsigned int cores)
static int
do_one_mempool_test(struct rte_mempool *mp, unsigned int cores, int external_cache)
{
+ unsigned int bulk_tab_max[] = { CACHE_LINE_BURST, 32, 64, 128, 256,
+ RTE_MEMPOOL_CACHE_MAX_SIZE, 0 };
unsigned int bulk_tab_get[] = { 1, 4, CACHE_LINE_BURST, 32, 64, 128, 256,
RTE_MEMPOOL_CACHE_MAX_SIZE, 0 };
unsigned int bulk_tab_put[] = { 1, 4, CACHE_LINE_BURST, 32, 64, 128, 256,
RTE_MEMPOOL_CACHE_MAX_SIZE, 0 };
unsigned int keep_tab[] = { 32, 128, 512, 2048, 8192, 32768, 0 };
+ unsigned int *max_bulk_ptr;
unsigned int *get_bulk_ptr;
unsigned int *put_bulk_ptr;
unsigned int *keep_ptr;
int ret;
+ for (keep_ptr = keep_tab; *keep_ptr; keep_ptr++) {
+ for (max_bulk_ptr = bulk_tab_max; *max_bulk_ptr; max_bulk_ptr++) {
+
+ if (*keep_ptr < *max_bulk_ptr)
+ continue;
+
+ use_external_cache = external_cache;
+ use_constant_values = 0;
+ n_max_bulk = *max_bulk_ptr;
+ n_get_bulk = 0;
+ n_put_bulk = 0;
+ n_keep = *keep_ptr;
+ ret = launch_cores(mp, cores);
+ if (ret < 0)
+ return -1;
+ }
+ }
+
for (keep_ptr = keep_tab; *keep_ptr; keep_ptr++) {
for (get_bulk_ptr = bulk_tab_get; *get_bulk_ptr; get_bulk_ptr++) {
for (put_bulk_ptr = bulk_tab_put; *put_bulk_ptr; put_bulk_ptr++) {
@@ -331,6 +417,7 @@ do_one_mempool_test(struct rte_mempool *mp, unsigned int cores, int external_cac
use_external_cache = external_cache;
use_constant_values = 0;
+ n_max_bulk = 0;
n_get_bulk = *get_bulk_ptr;
n_put_bulk = *put_bulk_ptr;
n_keep = *keep_ptr;
@@ -348,6 +435,7 @@ do_one_mempool_test(struct rte_mempool *mp, unsigned int cores, int external_cac
}
}
}
+
return 0;
}
--
2.43.0