test_ring_perf.c revision 9ecc306d
1/*-
2 *   BSD LICENSE
3 *
4 *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
5 *   All rights reserved.
6 *
7 *   Redistribution and use in source and binary forms, with or without
8 *   modification, are permitted provided that the following conditions
9 *   are met:
10 *
11 *     * Redistributions of source code must retain the above copyright
12 *       notice, this list of conditions and the following disclaimer.
13 *     * Redistributions in binary form must reproduce the above copyright
14 *       notice, this list of conditions and the following disclaimer in
15 *       the documentation and/or other materials provided with the
16 *       distribution.
17 *     * Neither the name of Intel Corporation nor the names of its
18 *       contributors may be used to endorse or promote products derived
19 *       from this software without specific prior written permission.
20 *
21 *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 */
33
34
35#include <stdio.h>
36#include <inttypes.h>
37#include <rte_ring.h>
38#include <rte_cycles.h>
39#include <rte_launch.h>
40
41#include "test.h"
42
43/*
44 * Ring
45 * ====
46 *
47 * Measures performance of various operations using rdtsc
48 *  * Empty ring dequeue
49 *  * Enqueue/dequeue of bursts in 1 threads
50 *  * Enqueue/dequeue of bursts in 2 threads
51 */
52
53#define RING_NAME "RING_PERF"
54#define RING_SIZE 4096
55#define MAX_BURST 32
56
57/*
58 * the sizes to enqueue and dequeue in testing
59 * (marked volatile so they won't be seen as compile-time constants)
60 */
61static const volatile unsigned bulk_sizes[] = { 8, 32 };
62
63/* The ring structure used for tests */
64static struct rte_ring *r;
65
66struct lcore_pair {
67	unsigned c1, c2;
68};
69
70static volatile unsigned lcore_count = 0;
71
72/**** Functions to analyse our core mask to get cores for different tests ***/
73
74static int
75get_two_hyperthreads(struct lcore_pair *lcp)
76{
77	unsigned id1, id2;
78	unsigned c1, c2, s1, s2;
79	RTE_LCORE_FOREACH(id1) {
80		/* inner loop just re-reads all id's. We could skip the first few
81		 * elements, but since number of cores is small there is little point
82		 */
83		RTE_LCORE_FOREACH(id2) {
84			if (id1 == id2)
85				continue;
86			c1 = lcore_config[id1].core_id;
87			c2 = lcore_config[id2].core_id;
88			s1 = lcore_config[id1].socket_id;
89			s2 = lcore_config[id2].socket_id;
90			if ((c1 == c2) && (s1 == s2)){
91				lcp->c1 = id1;
92				lcp->c2 = id2;
93				return 0;
94			}
95		}
96	}
97	return 1;
98}
99
100static int
101get_two_cores(struct lcore_pair *lcp)
102{
103	unsigned id1, id2;
104	unsigned c1, c2, s1, s2;
105	RTE_LCORE_FOREACH(id1) {
106		RTE_LCORE_FOREACH(id2) {
107			if (id1 == id2)
108				continue;
109			c1 = lcore_config[id1].core_id;
110			c2 = lcore_config[id2].core_id;
111			s1 = lcore_config[id1].socket_id;
112			s2 = lcore_config[id2].socket_id;
113			if ((c1 != c2) && (s1 == s2)){
114				lcp->c1 = id1;
115				lcp->c2 = id2;
116				return 0;
117			}
118		}
119	}
120	return 1;
121}
122
123static int
124get_two_sockets(struct lcore_pair *lcp)
125{
126	unsigned id1, id2;
127	unsigned s1, s2;
128	RTE_LCORE_FOREACH(id1) {
129		RTE_LCORE_FOREACH(id2) {
130			if (id1 == id2)
131				continue;
132			s1 = lcore_config[id1].socket_id;
133			s2 = lcore_config[id2].socket_id;
134			if (s1 != s2){
135				lcp->c1 = id1;
136				lcp->c2 = id2;
137				return 0;
138			}
139		}
140	}
141	return 1;
142}
143
144/* Get cycle counts for dequeuing from an empty ring. Should be 2 or 3 cycles */
145static void
146test_empty_dequeue(void)
147{
148	const unsigned iter_shift = 26;
149	const unsigned iterations = 1<<iter_shift;
150	unsigned i = 0;
151	void *burst[MAX_BURST];
152
153	const uint64_t sc_start = rte_rdtsc();
154	for (i = 0; i < iterations; i++)
155		rte_ring_sc_dequeue_bulk(r, burst, bulk_sizes[0]);
156	const uint64_t sc_end = rte_rdtsc();
157
158	const uint64_t mc_start = rte_rdtsc();
159	for (i = 0; i < iterations; i++)
160		rte_ring_mc_dequeue_bulk(r, burst, bulk_sizes[0]);
161	const uint64_t mc_end = rte_rdtsc();
162
163	printf("SC empty dequeue: %.2F\n",
164			(double)(sc_end-sc_start) / iterations);
165	printf("MC empty dequeue: %.2F\n",
166			(double)(mc_end-mc_start) / iterations);
167}
168
169/*
170 * for the separate enqueue and dequeue threads they take in one param
171 * and return two. Input = burst size, output = cycle average for sp/sc & mp/mc
172 */
173struct thread_params {
174	unsigned size;        /* input value, the burst size */
175	double spsc, mpmc;    /* output value, the single or multi timings */
176};
177
178/*
179 * Function that uses rdtsc to measure timing for ring enqueue. Needs pair
180 * thread running dequeue_bulk function
181 */
182static int
183enqueue_bulk(void *p)
184{
185	const unsigned iter_shift = 23;
186	const unsigned iterations = 1<<iter_shift;
187	struct thread_params *params = p;
188	const unsigned size = params->size;
189	unsigned i;
190	void *burst[MAX_BURST] = {0};
191
192	if ( __sync_add_and_fetch(&lcore_count, 1) != 2 )
193		while(lcore_count != 2)
194			rte_pause();
195
196	const uint64_t sp_start = rte_rdtsc();
197	for (i = 0; i < iterations; i++)
198		while (rte_ring_sp_enqueue_bulk(r, burst, size) != 0)
199			rte_pause();
200	const uint64_t sp_end = rte_rdtsc();
201
202	const uint64_t mp_start = rte_rdtsc();
203	for (i = 0; i < iterations; i++)
204		while (rte_ring_mp_enqueue_bulk(r, burst, size) != 0)
205			rte_pause();
206	const uint64_t mp_end = rte_rdtsc();
207
208	params->spsc = ((double)(sp_end - sp_start))/(iterations*size);
209	params->mpmc = ((double)(mp_end - mp_start))/(iterations*size);
210	return 0;
211}
212
213/*
214 * Function that uses rdtsc to measure timing for ring dequeue. Needs pair
215 * thread running enqueue_bulk function
216 */
217static int
218dequeue_bulk(void *p)
219{
220	const unsigned iter_shift = 23;
221	const unsigned iterations = 1<<iter_shift;
222	struct thread_params *params = p;
223	const unsigned size = params->size;
224	unsigned i;
225	void *burst[MAX_BURST] = {0};
226
227	if ( __sync_add_and_fetch(&lcore_count, 1) != 2 )
228		while(lcore_count != 2)
229			rte_pause();
230
231	const uint64_t sc_start = rte_rdtsc();
232	for (i = 0; i < iterations; i++)
233		while (rte_ring_sc_dequeue_bulk(r, burst, size) != 0)
234			rte_pause();
235	const uint64_t sc_end = rte_rdtsc();
236
237	const uint64_t mc_start = rte_rdtsc();
238	for (i = 0; i < iterations; i++)
239		while (rte_ring_mc_dequeue_bulk(r, burst, size) != 0)
240			rte_pause();
241	const uint64_t mc_end = rte_rdtsc();
242
243	params->spsc = ((double)(sc_end - sc_start))/(iterations*size);
244	params->mpmc = ((double)(mc_end - mc_start))/(iterations*size);
245	return 0;
246}
247
248/*
249 * Function that calls the enqueue and dequeue bulk functions on pairs of cores.
250 * used to measure ring perf between hyperthreads, cores and sockets.
251 */
252static void
253run_on_core_pair(struct lcore_pair *cores,
254		lcore_function_t f1, lcore_function_t f2)
255{
256	struct thread_params param1 = {0}, param2 = {0};
257	unsigned i;
258	for (i = 0; i < sizeof(bulk_sizes)/sizeof(bulk_sizes[0]); i++) {
259		lcore_count = 0;
260		param1.size = param2.size = bulk_sizes[i];
261		if (cores->c1 == rte_get_master_lcore()) {
262			rte_eal_remote_launch(f2, &param2, cores->c2);
263			f1(&param1);
264			rte_eal_wait_lcore(cores->c2);
265		} else {
266			rte_eal_remote_launch(f1, &param1, cores->c1);
267			rte_eal_remote_launch(f2, &param2, cores->c2);
268			rte_eal_wait_lcore(cores->c1);
269			rte_eal_wait_lcore(cores->c2);
270		}
271		printf("SP/SC bulk enq/dequeue (size: %u): %.2F\n", bulk_sizes[i],
272				param1.spsc + param2.spsc);
273		printf("MP/MC bulk enq/dequeue (size: %u): %.2F\n", bulk_sizes[i],
274				param1.mpmc + param2.mpmc);
275	}
276}
277
278/*
279 * Test function that determines how long an enqueue + dequeue of a single item
280 * takes on a single lcore. Result is for comparison with the bulk enq+deq.
281 */
282static void
283test_single_enqueue_dequeue(void)
284{
285	const unsigned iter_shift = 24;
286	const unsigned iterations = 1<<iter_shift;
287	unsigned i = 0;
288	void *burst = NULL;
289
290	const uint64_t sc_start = rte_rdtsc();
291	for (i = 0; i < iterations; i++) {
292		rte_ring_sp_enqueue(r, burst);
293		rte_ring_sc_dequeue(r, &burst);
294	}
295	const uint64_t sc_end = rte_rdtsc();
296
297	const uint64_t mc_start = rte_rdtsc();
298	for (i = 0; i < iterations; i++) {
299		rte_ring_mp_enqueue(r, burst);
300		rte_ring_mc_dequeue(r, &burst);
301	}
302	const uint64_t mc_end = rte_rdtsc();
303
304	printf("SP/SC single enq/dequeue: %"PRIu64"\n",
305			(sc_end-sc_start) >> iter_shift);
306	printf("MP/MC single enq/dequeue: %"PRIu64"\n",
307			(mc_end-mc_start) >> iter_shift);
308}
309
310/*
311 * Test that does both enqueue and dequeue on a core using the burst() API calls
312 * instead of the bulk() calls used in other tests. Results should be the same
313 * as for the bulk function called on a single lcore.
314 */
315static void
316test_burst_enqueue_dequeue(void)
317{
318	const unsigned iter_shift = 23;
319	const unsigned iterations = 1<<iter_shift;
320	unsigned sz, i = 0;
321	void *burst[MAX_BURST] = {0};
322
323	for (sz = 0; sz < sizeof(bulk_sizes)/sizeof(bulk_sizes[0]); sz++) {
324		const uint64_t sc_start = rte_rdtsc();
325		for (i = 0; i < iterations; i++) {
326			rte_ring_sp_enqueue_burst(r, burst, bulk_sizes[sz]);
327			rte_ring_sc_dequeue_burst(r, burst, bulk_sizes[sz]);
328		}
329		const uint64_t sc_end = rte_rdtsc();
330
331		const uint64_t mc_start = rte_rdtsc();
332		for (i = 0; i < iterations; i++) {
333			rte_ring_mp_enqueue_burst(r, burst, bulk_sizes[sz]);
334			rte_ring_mc_dequeue_burst(r, burst, bulk_sizes[sz]);
335		}
336		const uint64_t mc_end = rte_rdtsc();
337
338		uint64_t mc_avg = ((mc_end-mc_start) >> iter_shift) / bulk_sizes[sz];
339		uint64_t sc_avg = ((sc_end-sc_start) >> iter_shift) / bulk_sizes[sz];
340
341		printf("SP/SC burst enq/dequeue (size: %u): %"PRIu64"\n", bulk_sizes[sz],
342				sc_avg);
343		printf("MP/MC burst enq/dequeue (size: %u): %"PRIu64"\n", bulk_sizes[sz],
344				mc_avg);
345	}
346}
347
348/* Times enqueue and dequeue on a single lcore */
349static void
350test_bulk_enqueue_dequeue(void)
351{
352	const unsigned iter_shift = 23;
353	const unsigned iterations = 1<<iter_shift;
354	unsigned sz, i = 0;
355	void *burst[MAX_BURST] = {0};
356
357	for (sz = 0; sz < sizeof(bulk_sizes)/sizeof(bulk_sizes[0]); sz++) {
358		const uint64_t sc_start = rte_rdtsc();
359		for (i = 0; i < iterations; i++) {
360			rte_ring_sp_enqueue_bulk(r, burst, bulk_sizes[sz]);
361			rte_ring_sc_dequeue_bulk(r, burst, bulk_sizes[sz]);
362		}
363		const uint64_t sc_end = rte_rdtsc();
364
365		const uint64_t mc_start = rte_rdtsc();
366		for (i = 0; i < iterations; i++) {
367			rte_ring_mp_enqueue_bulk(r, burst, bulk_sizes[sz]);
368			rte_ring_mc_dequeue_bulk(r, burst, bulk_sizes[sz]);
369		}
370		const uint64_t mc_end = rte_rdtsc();
371
372		double sc_avg = ((double)(sc_end-sc_start) /
373				(iterations * bulk_sizes[sz]));
374		double mc_avg = ((double)(mc_end-mc_start) /
375				(iterations * bulk_sizes[sz]));
376
377		printf("SP/SC bulk enq/dequeue (size: %u): %.2F\n", bulk_sizes[sz],
378				sc_avg);
379		printf("MP/MC bulk enq/dequeue (size: %u): %.2F\n", bulk_sizes[sz],
380				mc_avg);
381	}
382}
383
384static int
385test_ring_perf(void)
386{
387	struct lcore_pair cores;
388	r = rte_ring_create(RING_NAME, RING_SIZE, rte_socket_id(), 0);
389	if (r == NULL && (r = rte_ring_lookup(RING_NAME)) == NULL)
390		return -1;
391
392	printf("### Testing single element and burst enq/deq ###\n");
393	test_single_enqueue_dequeue();
394	test_burst_enqueue_dequeue();
395
396	printf("\n### Testing empty dequeue ###\n");
397	test_empty_dequeue();
398
399	printf("\n### Testing using a single lcore ###\n");
400	test_bulk_enqueue_dequeue();
401
402	if (get_two_hyperthreads(&cores) == 0) {
403		printf("\n### Testing using two hyperthreads ###\n");
404		run_on_core_pair(&cores, enqueue_bulk, dequeue_bulk);
405	}
406	if (get_two_cores(&cores) == 0) {
407		printf("\n### Testing using two physical cores ###\n");
408		run_on_core_pair(&cores, enqueue_bulk, dequeue_bulk);
409	}
410	if (get_two_sockets(&cores) == 0) {
411		printf("\n### Testing using two NUMA nodes ###\n");
412		run_on_core_pair(&cores, enqueue_bulk, dequeue_bulk);
413	}
414	return 0;
415}
416
417REGISTER_TEST_COMMAND(ring_perf_autotest, test_ring_perf);
418