]> git.proxmox.com Git - ceph.git/blob - ceph/src/dpdk/app/test/test_lpm_perf.c
add subtree-ish sources for 12.0.3
[ceph.git] / ceph / src / dpdk / app / test / test_lpm_perf.c
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 #include <stdio.h>
35 #include <stdint.h>
36 #include <stdlib.h>
37 #include <math.h>
38
39 #include <rte_cycles.h>
40 #include <rte_random.h>
41 #include <rte_branch_prediction.h>
42 #include <rte_ip.h>
43 #include <rte_lpm.h>
44
45 #include "test.h"
46 #include "test_xmmt_ops.h"
47
48 #define TEST_LPM_ASSERT(cond) do { \
49 if (!(cond)) { \
50 printf("Error at line %d: \n", __LINE__); \
51 return -1; \
52 } \
53 } while(0)
54
55 #define ITERATIONS (1 << 10)
56 #define BATCH_SIZE (1 << 12)
57 #define BULK_SIZE 32
58
59 #define MAX_RULE_NUM (1200000)
60
61 struct route_rule {
62 uint32_t ip;
63 uint8_t depth;
64 };
65
66 struct route_rule large_route_table[MAX_RULE_NUM];
67
68 static uint32_t num_route_entries;
69 #define NUM_ROUTE_ENTRIES num_route_entries
70
71 enum {
72 IP_CLASS_A,
73 IP_CLASS_B,
74 IP_CLASS_C
75 };
76
77 /* struct route_rule_count defines the total number of rules in following a/b/c
78 * each item in a[]/b[]/c[] is the number of common IP address class A/B/C, not
79 * including the ones for private local network.
80 */
81 struct route_rule_count {
82 uint32_t a[RTE_LPM_MAX_DEPTH];
83 uint32_t b[RTE_LPM_MAX_DEPTH];
84 uint32_t c[RTE_LPM_MAX_DEPTH];
85 };
86
87 /* All following numbers of each depth of each common IP class are just
88 * got from previous large constant table in app/test/test_lpm_routes.h .
89 * In order to match similar performance, they keep same depth and IP
90 * address coverage as previous constant table. These numbers don't
91 * include any private local IP address. As previous large const rule
92 * table was just dumped from a real router, there are no any IP address
93 * in class C or D.
94 */
95 static struct route_rule_count rule_count = {
96 .a = { /* IP class A in which the most significant bit is 0 */
97 0, /* depth = 1 */
98 0, /* depth = 2 */
99 1, /* depth = 3 */
100 0, /* depth = 4 */
101 2, /* depth = 5 */
102 1, /* depth = 6 */
103 3, /* depth = 7 */
104 185, /* depth = 8 */
105 26, /* depth = 9 */
106 16, /* depth = 10 */
107 39, /* depth = 11 */
108 144, /* depth = 12 */
109 233, /* depth = 13 */
110 528, /* depth = 14 */
111 866, /* depth = 15 */
112 3856, /* depth = 16 */
113 3268, /* depth = 17 */
114 5662, /* depth = 18 */
115 17301, /* depth = 19 */
116 22226, /* depth = 20 */
117 11147, /* depth = 21 */
118 16746, /* depth = 22 */
119 17120, /* depth = 23 */
120 77578, /* depth = 24 */
121 401, /* depth = 25 */
122 656, /* depth = 26 */
123 1107, /* depth = 27 */
124 1121, /* depth = 28 */
125 2316, /* depth = 29 */
126 717, /* depth = 30 */
127 10, /* depth = 31 */
128 66 /* depth = 32 */
129 },
130 .b = { /* IP class A in which the most 2 significant bits are 10 */
131 0, /* depth = 1 */
132 0, /* depth = 2 */
133 0, /* depth = 3 */
134 0, /* depth = 4 */
135 1, /* depth = 5 */
136 1, /* depth = 6 */
137 1, /* depth = 7 */
138 3, /* depth = 8 */
139 3, /* depth = 9 */
140 30, /* depth = 10 */
141 25, /* depth = 11 */
142 168, /* depth = 12 */
143 305, /* depth = 13 */
144 569, /* depth = 14 */
145 1129, /* depth = 15 */
146 50800, /* depth = 16 */
147 1645, /* depth = 17 */
148 1820, /* depth = 18 */
149 3506, /* depth = 19 */
150 3258, /* depth = 20 */
151 3424, /* depth = 21 */
152 4971, /* depth = 22 */
153 6885, /* depth = 23 */
154 39771, /* depth = 24 */
155 424, /* depth = 25 */
156 170, /* depth = 26 */
157 433, /* depth = 27 */
158 92, /* depth = 28 */
159 366, /* depth = 29 */
160 377, /* depth = 30 */
161 2, /* depth = 31 */
162 200 /* depth = 32 */
163 },
164 .c = { /* IP class A in which the most 3 significant bits are 110 */
165 0, /* depth = 1 */
166 0, /* depth = 2 */
167 0, /* depth = 3 */
168 0, /* depth = 4 */
169 0, /* depth = 5 */
170 0, /* depth = 6 */
171 0, /* depth = 7 */
172 12, /* depth = 8 */
173 8, /* depth = 9 */
174 9, /* depth = 10 */
175 33, /* depth = 11 */
176 69, /* depth = 12 */
177 237, /* depth = 13 */
178 1007, /* depth = 14 */
179 1717, /* depth = 15 */
180 14663, /* depth = 16 */
181 8070, /* depth = 17 */
182 16185, /* depth = 18 */
183 48261, /* depth = 19 */
184 36870, /* depth = 20 */
185 33960, /* depth = 21 */
186 50638, /* depth = 22 */
187 61422, /* depth = 23 */
188 466549, /* depth = 24 */
189 1829, /* depth = 25 */
190 4824, /* depth = 26 */
191 4927, /* depth = 27 */
192 5914, /* depth = 28 */
193 10254, /* depth = 29 */
194 4905, /* depth = 30 */
195 1, /* depth = 31 */
196 716 /* depth = 32 */
197 }
198 };
199
200 static void generate_random_rule_prefix(uint32_t ip_class, uint8_t depth)
201 {
202 /* IP address class A, the most significant bit is 0 */
203 #define IP_HEAD_MASK_A 0x00000000
204 #define IP_HEAD_BIT_NUM_A 1
205
206 /* IP address class B, the most significant 2 bits are 10 */
207 #define IP_HEAD_MASK_B 0x80000000
208 #define IP_HEAD_BIT_NUM_B 2
209
210 /* IP address class C, the most significant 3 bits are 110 */
211 #define IP_HEAD_MASK_C 0xC0000000
212 #define IP_HEAD_BIT_NUM_C 3
213
214 uint32_t class_depth;
215 uint32_t range;
216 uint32_t mask;
217 uint32_t step;
218 uint32_t start;
219 uint32_t fixed_bit_num;
220 uint32_t ip_head_mask;
221 uint32_t rule_num;
222 uint32_t k;
223 struct route_rule *ptr_rule;
224
225 if (ip_class == IP_CLASS_A) { /* IP Address class A */
226 fixed_bit_num = IP_HEAD_BIT_NUM_A;
227 ip_head_mask = IP_HEAD_MASK_A;
228 rule_num = rule_count.a[depth - 1];
229 } else if (ip_class == IP_CLASS_B) { /* IP Address class B */
230 fixed_bit_num = IP_HEAD_BIT_NUM_B;
231 ip_head_mask = IP_HEAD_MASK_B;
232 rule_num = rule_count.b[depth - 1];
233 } else { /* IP Address class C */
234 fixed_bit_num = IP_HEAD_BIT_NUM_C;
235 ip_head_mask = IP_HEAD_MASK_C;
236 rule_num = rule_count.c[depth - 1];
237 }
238
239 if (rule_num == 0)
240 return;
241
242 /* the number of rest bits which don't include the most significant
243 * fixed bits for this IP address class
244 */
245 class_depth = depth - fixed_bit_num;
246
247 /* range is the maximum number of rules for this depth and
248 * this IP address class
249 */
250 range = 1 << class_depth;
251
252 /* only mask the most depth significant generated bits
253 * except fixed bits for IP address class
254 */
255 mask = range - 1;
256
257 /* Widen coverage of IP address in generated rules */
258 if (range <= rule_num)
259 step = 1;
260 else
261 step = round((double)range / rule_num);
262
263 /* Only generate rest bits except the most significant
264 * fixed bits for IP address class
265 */
266 start = lrand48() & mask;
267 ptr_rule = &large_route_table[num_route_entries];
268 for (k = 0; k < rule_num; k++) {
269 ptr_rule->ip = (start << (RTE_LPM_MAX_DEPTH - depth))
270 | ip_head_mask;
271 ptr_rule->depth = depth;
272 ptr_rule++;
273 start = (start + step) & mask;
274 }
275 num_route_entries += rule_num;
276 }
277
278 static void insert_rule_in_random_pos(uint32_t ip, uint8_t depth)
279 {
280 uint32_t pos;
281 int try_count = 0;
282 struct route_rule tmp;
283
284 do {
285 pos = lrand48();
286 try_count++;
287 } while ((try_count < 10) && (pos > num_route_entries));
288
289 if ((pos > num_route_entries) || (pos >= MAX_RULE_NUM))
290 pos = num_route_entries >> 1;
291
292 tmp = large_route_table[pos];
293 large_route_table[pos].ip = ip;
294 large_route_table[pos].depth = depth;
295 if (num_route_entries < MAX_RULE_NUM)
296 large_route_table[num_route_entries++] = tmp;
297 }
298
299 static void generate_large_route_rule_table(void)
300 {
301 uint32_t ip_class;
302 uint8_t depth;
303
304 num_route_entries = 0;
305 memset(large_route_table, 0, sizeof(large_route_table));
306
307 for (ip_class = IP_CLASS_A; ip_class <= IP_CLASS_C; ip_class++) {
308 for (depth = 1; depth <= RTE_LPM_MAX_DEPTH; depth++) {
309 generate_random_rule_prefix(ip_class, depth);
310 }
311 }
312
313 /* Add following rules to keep same as previous large constant table,
314 * they are 4 rules with private local IP address and 1 all-zeros prefix
315 * with depth = 8.
316 */
317 insert_rule_in_random_pos(IPv4(0, 0, 0, 0), 8);
318 insert_rule_in_random_pos(IPv4(10, 2, 23, 147), 32);
319 insert_rule_in_random_pos(IPv4(192, 168, 100, 10), 24);
320 insert_rule_in_random_pos(IPv4(192, 168, 25, 100), 24);
321 insert_rule_in_random_pos(IPv4(192, 168, 129, 124), 32);
322 }
323
324 static void
325 print_route_distribution(const struct route_rule *table, uint32_t n)
326 {
327 unsigned i, j;
328
329 printf("Route distribution per prefix width: \n");
330 printf("DEPTH QUANTITY (PERCENT)\n");
331 printf("--------------------------- \n");
332
333 /* Count depths. */
334 for (i = 1; i <= 32; i++) {
335 unsigned depth_counter = 0;
336 double percent_hits;
337
338 for (j = 0; j < n; j++)
339 if (table[j].depth == (uint8_t) i)
340 depth_counter++;
341
342 percent_hits = ((double)depth_counter)/((double)n) * 100;
343 printf("%.2u%15u (%.2f)\n", i, depth_counter, percent_hits);
344 }
345 printf("\n");
346 }
347
348 static int
349 test_lpm_perf(void)
350 {
351 struct rte_lpm *lpm = NULL;
352 struct rte_lpm_config config;
353
354 config.max_rules = 2000000;
355 config.number_tbl8s = 2048;
356 config.flags = 0;
357 uint64_t begin, total_time, lpm_used_entries = 0;
358 unsigned i, j;
359 uint32_t next_hop_add = 0xAA, next_hop_return = 0;
360 int status = 0;
361 uint64_t cache_line_counter = 0;
362 int64_t count = 0;
363
364 rte_srand(rte_rdtsc());
365
366 generate_large_route_rule_table();
367
368 printf("No. routes = %u\n", (unsigned) NUM_ROUTE_ENTRIES);
369
370 print_route_distribution(large_route_table, (uint32_t) NUM_ROUTE_ENTRIES);
371
372 lpm = rte_lpm_create(__func__, SOCKET_ID_ANY, &config);
373 TEST_LPM_ASSERT(lpm != NULL);
374
375 /* Measue add. */
376 begin = rte_rdtsc();
377
378 for (i = 0; i < NUM_ROUTE_ENTRIES; i++) {
379 if (rte_lpm_add(lpm, large_route_table[i].ip,
380 large_route_table[i].depth, next_hop_add) == 0)
381 status++;
382 }
383 /* End Timer. */
384 total_time = rte_rdtsc() - begin;
385
386 printf("Unique added entries = %d\n", status);
387 /* Obtain add statistics. */
388 for (i = 0; i < RTE_LPM_TBL24_NUM_ENTRIES; i++) {
389 if (lpm->tbl24[i].valid)
390 lpm_used_entries++;
391
392 if (i % 32 == 0) {
393 if ((uint64_t)count < lpm_used_entries) {
394 cache_line_counter++;
395 count = lpm_used_entries;
396 }
397 }
398 }
399
400 printf("Used table 24 entries = %u (%g%%)\n",
401 (unsigned) lpm_used_entries,
402 (lpm_used_entries * 100.0) / RTE_LPM_TBL24_NUM_ENTRIES);
403 printf("64 byte Cache entries used = %u (%u bytes)\n",
404 (unsigned) cache_line_counter, (unsigned) cache_line_counter * 64);
405
406 printf("Average LPM Add: %g cycles\n",
407 (double)total_time / NUM_ROUTE_ENTRIES);
408
409 /* Measure single Lookup */
410 total_time = 0;
411 count = 0;
412
413 for (i = 0; i < ITERATIONS; i++) {
414 static uint32_t ip_batch[BATCH_SIZE];
415
416 for (j = 0; j < BATCH_SIZE; j++)
417 ip_batch[j] = rte_rand();
418
419 /* Lookup per batch */
420 begin = rte_rdtsc();
421
422 for (j = 0; j < BATCH_SIZE; j++) {
423 if (rte_lpm_lookup(lpm, ip_batch[j], &next_hop_return) != 0)
424 count++;
425 }
426
427 total_time += rte_rdtsc() - begin;
428
429 }
430 printf("Average LPM Lookup: %.1f cycles (fails = %.1f%%)\n",
431 (double)total_time / ((double)ITERATIONS * BATCH_SIZE),
432 (count * 100.0) / (double)(ITERATIONS * BATCH_SIZE));
433
434 /* Measure bulk Lookup */
435 total_time = 0;
436 count = 0;
437 for (i = 0; i < ITERATIONS; i++) {
438 static uint32_t ip_batch[BATCH_SIZE];
439 uint32_t next_hops[BULK_SIZE];
440
441 /* Create array of random IP addresses */
442 for (j = 0; j < BATCH_SIZE; j++)
443 ip_batch[j] = rte_rand();
444
445 /* Lookup per batch */
446 begin = rte_rdtsc();
447 for (j = 0; j < BATCH_SIZE; j += BULK_SIZE) {
448 unsigned k;
449 rte_lpm_lookup_bulk(lpm, &ip_batch[j], next_hops, BULK_SIZE);
450 for (k = 0; k < BULK_SIZE; k++)
451 if (unlikely(!(next_hops[k] & RTE_LPM_LOOKUP_SUCCESS)))
452 count++;
453 }
454
455 total_time += rte_rdtsc() - begin;
456 }
457 printf("BULK LPM Lookup: %.1f cycles (fails = %.1f%%)\n",
458 (double)total_time / ((double)ITERATIONS * BATCH_SIZE),
459 (count * 100.0) / (double)(ITERATIONS * BATCH_SIZE));
460
461 /* Measure LookupX4 */
462 total_time = 0;
463 count = 0;
464 for (i = 0; i < ITERATIONS; i++) {
465 static uint32_t ip_batch[BATCH_SIZE];
466 uint32_t next_hops[4];
467
468 /* Create array of random IP addresses */
469 for (j = 0; j < BATCH_SIZE; j++)
470 ip_batch[j] = rte_rand();
471
472 /* Lookup per batch */
473 begin = rte_rdtsc();
474 for (j = 0; j < BATCH_SIZE; j += RTE_DIM(next_hops)) {
475 unsigned k;
476 xmm_t ipx4;
477
478 ipx4 = vect_loadu_sil128((xmm_t *)(ip_batch + j));
479 ipx4 = *(xmm_t *)(ip_batch + j);
480 rte_lpm_lookupx4(lpm, ipx4, next_hops, UINT32_MAX);
481 for (k = 0; k < RTE_DIM(next_hops); k++)
482 if (unlikely(next_hops[k] == UINT32_MAX))
483 count++;
484 }
485
486 total_time += rte_rdtsc() - begin;
487 }
488 printf("LPM LookupX4: %.1f cycles (fails = %.1f%%)\n",
489 (double)total_time / ((double)ITERATIONS * BATCH_SIZE),
490 (count * 100.0) / (double)(ITERATIONS * BATCH_SIZE));
491
492 /* Delete */
493 status = 0;
494 begin = rte_rdtsc();
495
496 for (i = 0; i < NUM_ROUTE_ENTRIES; i++) {
497 /* rte_lpm_delete(lpm, ip, depth) */
498 status += rte_lpm_delete(lpm, large_route_table[i].ip,
499 large_route_table[i].depth);
500 }
501
502 total_time += rte_rdtsc() - begin;
503
504 printf("Average LPM Delete: %g cycles\n",
505 (double)total_time / NUM_ROUTE_ENTRIES);
506
507 rte_lpm_delete_all(lpm);
508 rte_lpm_free(lpm);
509
510 return 0;
511 }
512
513 REGISTER_TEST_COMMAND(lpm_perf_autotest, test_lpm_perf);