1.
해외 기술논문을 읽어보면 Ring Buffer를 이용한 기술들이 많습니다. LMAX의 Disruptor도 역시 Ring Buffer를 이용하여 Multi-Threading Application을 구현하고 있습니다. 오늘 소개하는 논문도 Ring Buffer와 관련한 자료입니다.
논문을 알게된 계기는 저자가 대표로 있는 회사의 블로그입니다.Linux Journal에 실린 글입니다. 제목은 “How to Scale the Work Queue in a Multicore Environment”입니다. 전문을 함께 올립니다.
2.
아울러 논문에서 소개한 소스도 함께소개합니다.
Lock-Free Multi-Producer Multi-Consumer Queue on Ring Buffer
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 |
/** * Implementation of Naive and Lock-free ring-buffer queues and * performance and verification tests. * * Build with (g++ version must be >= 4.5.0): * $ g++ -Wall -std=c++0x -O2 -D DCACHE1_LINESIZE=`getconf LEVEL1_DCACHE_LINESIZE` lockfree_rb_q.cc -lpthread * * I verified the program with g++ 4.5.3, 4.6.1 and 4.6.3. * * Copyright (C) 2012-2013 Alexander Krizhanovsky (ak@natsys-lab.com). * * This file is free software; you can redistribute it and/or modify * it under the terms of the GNU Lesser General Public License as published * by the Free Software Foundation; either version 3, or (at your option) * any later version. * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU Lesser General Public License for more details. * See http://www.gnu.org/licenses/lgpl.html . */ #ifndef __x86_64__ #warning "The program is developed for x86-64 architecture only." #endif #if !defined(DCACHE1_LINESIZE) || !DCACHE1_LINESIZE #ifdef DCACHE1_LINESIZE #undef DCACHE1_LINESIZE #endif #define ____cacheline_aligned __attribute__((aligned(64))) #endif #define ____cacheline_aligned __attribute__((aligned(DCACHE1_LINESIZE))) #include <limits.h> #include <malloc.h> #include <string.h> #include <unistd.h> #include <atomic> #include <cassert> #include <iostream> #include <condition_variable> #include <mutex> #include <thread> #define QUEUE_SIZE (32 * 1024) /* * ------------------------------------------------------------------------ * Naive serialized ring buffer queue * ------------------------------------------------------------------------ */ template<class T, unsigned long Q_SIZE = QUEUE_SIZE> class NaiveQueue { private: static const unsigned long Q_MASK = Q_SIZE - 1; public: NaiveQueue() : head_(0), tail_(0) { ptr_array_ = (T **)::memalign(getpagesize(), Q_SIZE * sizeof(void *)); assert(ptr_array_); } void push(T *x) { std::unique_lock<std::mutex> lock(mtx_); cond_overflow_.wait(lock, [&head_, &tail_]() { return tail_ + Q_SIZE > head_; }); ptr_array_[head_++ & Q_MASK] = x; cond_empty_.notify_one(); } T * pop() { std::unique_lock<std::mutex> lock(mtx_); cond_empty_.wait(lock, [&head_, &tail_]() { return tail_ < head_; }); T *x = ptr_array_[tail_++ & Q_MASK]; cond_overflow_.notify_one(); return x; } private: unsigned long head_, tail_; std::condition_variable cond_empty_; std::condition_variable cond_overflow_; std::mutex mtx_; T **ptr_array_; }; /* * ------------------------------------------------------------------------ * Lock-free N-producers M-consumers ring-buffer queue. * ABA problem safe. * * This implementation is bit complicated, so possibly it has sense to use * classic list-based queues. See: * 1. D.Fober, Y.Orlarey, S.Letz, "Lock-Free Techniques for Concurrent * Access to Shared Ojects" * 2. M.M.Michael, M.L.Scott, "Simple, Fast and Practical Non-Blocking and * Blocking Concurrent Queue Algorithms" * 3. E.Ladan-Mozes, N.Shavit, "An Optimistic Approach to Lock-Free FIFO Queues" * * See also implementation of N-producers M-consumers FIFO and * 1-producer 1-consumer ring-buffer from Tim Blechmann: * http://tim.klingt.org/boost_lockfree/ * git://tim.klingt.org/boost_lockfree.git * * See See Intel 64 and IA-32 Architectures Software Developer's Manual, * Volume 3, Chapter 8.2 Memory Ordering for x86 memory ordering guarantees. * ------------------------------------------------------------------------ */ // thread_local is still not implemented in GCC. static size_t __thread __thr_id; /** * @return continous thread IDs starting from 0 as opposed to pthread_self(). */ inline size_t thr_id() { return __thr_id; } inline void set_thr_id(size_t id) { __thr_id = id; } template<class T, decltype(thr_id) ThrId = thr_id, unsigned long Q_SIZE = QUEUE_SIZE> class LockFreeQueue { private: static const unsigned long Q_MASK = Q_SIZE - 1; struct ThrPos { unsigned long head, tail; }; public: LockFreeQueue(size_t n_producers, size_t n_consumers) : n_producers_(n_producers), n_consumers_(n_consumers), head_(0), tail_(0), last_head_(0), last_tail_(0) { auto n = std::max(n_consumers_, n_producers_); thr_p_ = (ThrPos *)::memalign(getpagesize(), sizeof(ThrPos) * n); assert(thr_p_); // Set per thread tail and head to ULONG_MAX. ::memset((void *)thr_p_, 0xFF, sizeof(ThrPos) * n); ptr_array_ = (T **)::memalign(getpagesize(), Q_SIZE * sizeof(void *)); assert(ptr_array_); } ~LockFreeQueue() { ::free(ptr_array_); ::free(thr_p_); } ThrPos& thr_pos() const { assert(ThrId() < std::max(n_consumers_, n_producers_)); return thr_p_[ThrId()]; } void push(T *ptr) { /* * Request next place to push. * * Second assignemnt is atomic only for head shift, so there is * a time window in which thr_p_[tid].head = ULONG_MAX, and * head could be shifted significantly by other threads, * so pop() will set last_head_ to head. * After that thr_p_[tid].head is setted to old head value * (which is stored in local CPU register) and written by @ptr. * * First assignment guaranties that pop() sees values for * head and thr_p_[tid].head not greater that they will be * after the second assignment with head shift. * * Loads and stores are not reordered with locked instructions, * se we don't need a memory barrier here. */ thr_pos().head = head_; thr_pos().head = __sync_fetch_and_add(&head_, 1); /* * We do not know when a consumer uses the pop()'ed pointer, * se we can not overwrite it and have to wait the lowest tail. */ while (__builtin_expect(thr_pos().head >= last_tail_ + Q_SIZE, 0)) { ::sched_yield(); auto min = tail_; // Update the last_tail_. for (size_t i = 0; i < n_consumers_; ++i) { auto tmp_t = thr_p_[i].tail; // Force compiler to use tmp_h exactly once. asm volatile("" ::: "memory"); if (tmp_t < min) min = tmp_t; } last_tail_ = min; } ptr_array_[thr_pos().head & Q_MASK] = ptr; // Allow consumers eat the item. thr_pos().head = ULONG_MAX; } T * pop() { /* * Request next place from which to pop. * See comments for push(). * * Loads and stores are not reordered with locked instructions, * se we don't need a memory barrier here. */ thr_pos().tail = tail_; thr_pos().tail = __sync_fetch_and_add(&tail_, 1); /* * tid'th place in ptr_array_ is reserved by the thread - * this place shall never be rewritten by push() and * last_tail_ at push() is a guarantee. * last_head_ guaraties that no any consumer eats the item * before producer reserved the position writes to it. */ while (__builtin_expect(thr_pos().tail >= last_head_, 0)) { ::sched_yield(); auto min = head_; // Update the last_head_. for (size_t i = 0; i < n_producers_; ++i) { auto tmp_h = thr_p_[i].head; // Force compiler to use tmp_h exactly once. asm volatile("" ::: "memory"); if (tmp_h < min) min = tmp_h; } last_head_ = min; } T *ret = ptr_array_[thr_pos().tail & Q_MASK]; // Allow producers rewrite the slot. thr_pos().tail = ULONG_MAX; return ret; } private: /* * The most hot members are cacheline aligned to avoid * False Sharing. */ const size_t n_producers_, n_consumers_; // currently free position (next to insert) unsigned long head_ ____cacheline_aligned; // current tail, next to pop unsigned long tail_ ____cacheline_aligned; // last not-processed producer's pointer unsigned long last_head_ ____cacheline_aligned; // last not-processed consumer's pointer unsigned long last_tail_ ____cacheline_aligned; ThrPos *thr_p_; T **ptr_array_; }; /* * ------------------------------------------------------------------------ * Tests for naive and lock-free queues * ------------------------------------------------------------------------ */ static const auto N = QUEUE_SIZE * 1024; static const auto CONSUMERS = 16; static const auto PRODUCERS = 16; typedef unsigned char q_type; static const q_type X_EMPTY = 0; // the address skipped by producers static const q_type X_MISSED = 255; // the address skipped by consumers q_type x[N * PRODUCERS]; std::atomic<int> n(0); template<class Q> struct Worker { Worker(Q *q, size_t id = 0) : q_(q), thr_id_(id) {} Q *q_; size_t thr_id_; }; template<class Q> struct Producer : public Worker<Q> { Producer(Q *q, size_t id) : Worker<Q>(q, id) {} void operator()() { set_thr_id(Worker<Q>::thr_id_); for (auto i = thr_id(); i < N * PRODUCERS; i += PRODUCERS) { x[i] = X_MISSED; Worker<Q>::q_->push(x + i); } } }; template<class Q> struct Consumer : public Worker<Q> { Consumer(Q *q, size_t id) : Worker<Q>(q, id) {} void operator()() { set_thr_id(Worker<Q>::thr_id_); while (n.fetch_add(1) < N * PRODUCERS) { q_type *v = Worker<Q>::q_->pop(); assert(v); assert(*v == X_MISSED); *v = (q_type)(thr_id() + 1); // don't write zero } } }; template<class Q> void run_test(Q &&q) { std::thread thr[PRODUCERS + CONSUMERS]; n.store(0); ::memset(x, X_EMPTY, N * sizeof(q_type) * PRODUCERS); // Run producers. for (auto i = 0; i < PRODUCERS; ++i) thr[i] = std::thread(Producer<Q>(&q, i)); ::usleep(10 * 1000); // sleep to wait the queue is full /* * Run consumers. * Create consumers with the same thread IDs as producers. * The IDs are used for queue head and tail indexing only, * so we care only about different IDs for threads of the same type. */ for (auto i = 0; i < CONSUMERS; ++i) thr[PRODUCERS + i] = std::thread(Consumer<Q>(&q, i)); // Wait for all threads completion. for (auto i = 0; i < PRODUCERS + CONSUMERS; ++i) thr[i].join(); // Check data. auto res = 0; std::cout << "check X data..." << std::endl; for (auto i = 0; i < N * PRODUCERS; ++i) { if (x[i] == X_EMPTY) { std::cout << "empty " << i << std::endl; res = 1; break; } else if (x[i] == X_MISSED) { std::cout << "missed " << i << std::endl; res = 2; break; } } std::cout << (res ? "FAILED" : "Passed") << std::endl; } int main() { LockFreeQueue<q_type> lf_q(PRODUCERS, CONSUMERS); run_test<LockFreeQueue<q_type>>(std::move(lf_q)); NaiveQueue<q_type> n_q; run_test<NaiveQueue<q_type>>(std::move(n_q)); return 0; } |
해당 소스 RunTest를 해보려 하니 오류가 발생합니다.
운영체제 : Window 10
개발환경 : Visual Studio 2015
#include 해당 부분이 Linux 운영체제가 아니라 SourceForge같은 곳에 있는 대체코드로 넣는데 aligned나 DCACHE1_LINESIZE 부분에 선언이 되지않았다고 뜨네요.
혹시 Window 버전의 소스도 공유가 가능할까요?
제가 개발한 소스가 아닙니다. URL에서 원문을 참조하시길 바랍니다. 그리고 MPMC를 위한 라이브러리는 무척 많습니다. 2013년자료임을 참조하시길 바랍니다.