db_bench_tree_db.cc raw
1 // Copyright (c) 2011 The LevelDB Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file. See the AUTHORS file for names of contributors.
4
5 #include <kcpolydb.h>
6 #include <stdio.h>
7 #include <stdlib.h>
8
9 #include "util/histogram.h"
10 #include "util/random.h"
11 #include "util/testutil.h"
12
13 // Comma-separated list of operations to run in the specified order
14 // Actual benchmarks:
15 //
16 // fillseq -- write N values in sequential key order in async mode
17 // fillrandom -- write N values in random key order in async mode
18 // overwrite -- overwrite N values in random key order in async mode
19 // fillseqsync -- write N/100 values in sequential key order in sync mode
20 // fillrandsync -- write N/100 values in random key order in sync mode
21 // fillrand100K -- write N/1000 100K values in random order in async mode
22 // fillseq100K -- write N/1000 100K values in seq order in async mode
23 // readseq -- read N times sequentially
24 // readseq100K -- read N/1000 100K values in sequential order in async mode
25 // readrand100K -- read N/1000 100K values in sequential order in async mode
26 // readrandom -- read N times in random order
27 static const char* FLAGS_benchmarks =
28 "fillseq,"
29 "fillseqsync,"
30 "fillrandsync,"
31 "fillrandom,"
32 "overwrite,"
33 "readrandom,"
34 "readseq,"
35 "fillrand100K,"
36 "fillseq100K,"
37 "readseq100K,"
38 "readrand100K,";
39
40 // Number of key/values to place in database
41 static int FLAGS_num = 1000000;
42
43 // Number of read operations to do. If negative, do FLAGS_num reads.
44 static int FLAGS_reads = -1;
45
46 // Size of each value
47 static int FLAGS_value_size = 100;
48
49 // Arrange to generate values that shrink to this fraction of
50 // their original size after compression
51 static double FLAGS_compression_ratio = 0.5;
52
53 // Print histogram of operation timings
54 static bool FLAGS_histogram = false;
55
56 // Cache size. Default 4 MB
57 static int FLAGS_cache_size = 4194304;
58
59 // Page size. Default 1 KB
60 static int FLAGS_page_size = 1024;
61
62 // If true, do not destroy the existing database. If you set this
63 // flag and also specify a benchmark that wants a fresh database, that
64 // benchmark will fail.
65 static bool FLAGS_use_existing_db = false;
66
67 // Compression flag. If true, compression is on. If false, compression
68 // is off.
69 static bool FLAGS_compression = true;
70
71 // Use the db with the following name.
72 static const char* FLAGS_db = nullptr;
73
74 inline static void DBSynchronize(kyotocabinet::TreeDB* db_) {
75 // Synchronize will flush writes to disk
76 if (!db_->synchronize()) {
77 fprintf(stderr, "synchronize error: %s\n", db_->error().name());
78 }
79 }
80
81 namespace leveldb {
82
83 // Helper for quickly generating random data.
84 namespace {
85 class RandomGenerator {
86 private:
87 std::string data_;
88 int pos_;
89
90 public:
91 RandomGenerator() {
92 // We use a limited amount of data over and over again and ensure
93 // that it is larger than the compression window (32KB), and also
94 // large enough to serve all typical value sizes we want to write.
95 Random rnd(301);
96 std::string piece;
97 while (data_.size() < 1048576) {
98 // Add a short fragment that is as compressible as specified
99 // by FLAGS_compression_ratio.
100 test::CompressibleString(&rnd, FLAGS_compression_ratio, 100, &piece);
101 data_.append(piece);
102 }
103 pos_ = 0;
104 }
105
106 Slice Generate(int len) {
107 if (pos_ + len > data_.size()) {
108 pos_ = 0;
109 assert(len < data_.size());
110 }
111 pos_ += len;
112 return Slice(data_.data() + pos_ - len, len);
113 }
114 };
115
116 static Slice TrimSpace(Slice s) {
117 int start = 0;
118 while (start < s.size() && isspace(s[start])) {
119 start++;
120 }
121 int limit = s.size();
122 while (limit > start && isspace(s[limit - 1])) {
123 limit--;
124 }
125 return Slice(s.data() + start, limit - start);
126 }
127
128 } // namespace
129
130 class Benchmark {
131 private:
132 kyotocabinet::TreeDB* db_;
133 int db_num_;
134 int num_;
135 int reads_;
136 double start_;
137 double last_op_finish_;
138 int64_t bytes_;
139 std::string message_;
140 Histogram hist_;
141 RandomGenerator gen_;
142 Random rand_;
143 kyotocabinet::LZOCompressor<kyotocabinet::LZO::RAW> comp_;
144
145 // State kept for progress messages
146 int done_;
147 int next_report_; // When to report next
148
149 void PrintHeader() {
150 const int kKeySize = 16;
151 PrintEnvironment();
152 fprintf(stdout, "Keys: %d bytes each\n", kKeySize);
153 fprintf(stdout, "Values: %d bytes each (%d bytes after compression)\n",
154 FLAGS_value_size,
155 static_cast<int>(FLAGS_value_size * FLAGS_compression_ratio + 0.5));
156 fprintf(stdout, "Entries: %d\n", num_);
157 fprintf(stdout, "RawSize: %.1f MB (estimated)\n",
158 ((static_cast<int64_t>(kKeySize + FLAGS_value_size) * num_) /
159 1048576.0));
160 fprintf(stdout, "FileSize: %.1f MB (estimated)\n",
161 (((kKeySize + FLAGS_value_size * FLAGS_compression_ratio) * num_) /
162 1048576.0));
163 PrintWarnings();
164 fprintf(stdout, "------------------------------------------------\n");
165 }
166
167 void PrintWarnings() {
168 #if defined(__GNUC__) && !defined(__OPTIMIZE__)
169 fprintf(
170 stdout,
171 "WARNING: Optimization is disabled: benchmarks unnecessarily slow\n");
172 #endif
173 #ifndef NDEBUG
174 fprintf(stdout,
175 "WARNING: Assertions are enabled; benchmarks unnecessarily slow\n");
176 #endif
177 }
178
179 void PrintEnvironment() {
180 fprintf(stderr, "Kyoto Cabinet: version %s, lib ver %d, lib rev %d\n",
181 kyotocabinet::VERSION, kyotocabinet::LIBVER, kyotocabinet::LIBREV);
182
183 #if defined(__linux)
184 time_t now = time(nullptr);
185 fprintf(stderr, "Date: %s", ctime(&now)); // ctime() adds newline
186
187 FILE* cpuinfo = fopen("/proc/cpuinfo", "r");
188 if (cpuinfo != nullptr) {
189 char line[1000];
190 int num_cpus = 0;
191 std::string cpu_type;
192 std::string cache_size;
193 while (fgets(line, sizeof(line), cpuinfo) != nullptr) {
194 const char* sep = strchr(line, ':');
195 if (sep == nullptr) {
196 continue;
197 }
198 Slice key = TrimSpace(Slice(line, sep - 1 - line));
199 Slice val = TrimSpace(Slice(sep + 1));
200 if (key == "model name") {
201 ++num_cpus;
202 cpu_type = val.ToString();
203 } else if (key == "cache size") {
204 cache_size = val.ToString();
205 }
206 }
207 fclose(cpuinfo);
208 fprintf(stderr, "CPU: %d * %s\n", num_cpus, cpu_type.c_str());
209 fprintf(stderr, "CPUCache: %s\n", cache_size.c_str());
210 }
211 #endif
212 }
213
214 void Start() {
215 start_ = Env::Default()->NowMicros() * 1e-6;
216 bytes_ = 0;
217 message_.clear();
218 last_op_finish_ = start_;
219 hist_.Clear();
220 done_ = 0;
221 next_report_ = 100;
222 }
223
224 void FinishedSingleOp() {
225 if (FLAGS_histogram) {
226 double now = Env::Default()->NowMicros() * 1e-6;
227 double micros = (now - last_op_finish_) * 1e6;
228 hist_.Add(micros);
229 if (micros > 20000) {
230 fprintf(stderr, "long op: %.1f micros%30s\r", micros, "");
231 fflush(stderr);
232 }
233 last_op_finish_ = now;
234 }
235
236 done_++;
237 if (done_ >= next_report_) {
238 if (next_report_ < 1000)
239 next_report_ += 100;
240 else if (next_report_ < 5000)
241 next_report_ += 500;
242 else if (next_report_ < 10000)
243 next_report_ += 1000;
244 else if (next_report_ < 50000)
245 next_report_ += 5000;
246 else if (next_report_ < 100000)
247 next_report_ += 10000;
248 else if (next_report_ < 500000)
249 next_report_ += 50000;
250 else
251 next_report_ += 100000;
252 fprintf(stderr, "... finished %d ops%30s\r", done_, "");
253 fflush(stderr);
254 }
255 }
256
257 void Stop(const Slice& name) {
258 double finish = Env::Default()->NowMicros() * 1e-6;
259
260 // Pretend at least one op was done in case we are running a benchmark
261 // that does not call FinishedSingleOp().
262 if (done_ < 1) done_ = 1;
263
264 if (bytes_ > 0) {
265 char rate[100];
266 snprintf(rate, sizeof(rate), "%6.1f MB/s",
267 (bytes_ / 1048576.0) / (finish - start_));
268 if (!message_.empty()) {
269 message_ = std::string(rate) + " " + message_;
270 } else {
271 message_ = rate;
272 }
273 }
274
275 fprintf(stdout, "%-12s : %11.3f micros/op;%s%s\n", name.ToString().c_str(),
276 (finish - start_) * 1e6 / done_, (message_.empty() ? "" : " "),
277 message_.c_str());
278 if (FLAGS_histogram) {
279 fprintf(stdout, "Microseconds per op:\n%s\n", hist_.ToString().c_str());
280 }
281 fflush(stdout);
282 }
283
284 public:
285 enum Order { SEQUENTIAL, RANDOM };
286 enum DBState { FRESH, EXISTING };
287
288 Benchmark()
289 : db_(nullptr),
290 num_(FLAGS_num),
291 reads_(FLAGS_reads < 0 ? FLAGS_num : FLAGS_reads),
292 bytes_(0),
293 rand_(301) {
294 std::vector<std::string> files;
295 std::string test_dir;
296 Env::Default()->GetTestDirectory(&test_dir);
297 Env::Default()->GetChildren(test_dir.c_str(), &files);
298 if (!FLAGS_use_existing_db) {
299 for (int i = 0; i < files.size(); i++) {
300 if (Slice(files[i]).starts_with("dbbench_polyDB")) {
301 std::string file_name(test_dir);
302 file_name += "/";
303 file_name += files[i];
304 Env::Default()->DeleteFile(file_name.c_str());
305 }
306 }
307 }
308 }
309
310 ~Benchmark() {
311 if (!db_->close()) {
312 fprintf(stderr, "close error: %s\n", db_->error().name());
313 }
314 }
315
316 void Run() {
317 PrintHeader();
318 Open(false);
319
320 const char* benchmarks = FLAGS_benchmarks;
321 while (benchmarks != nullptr) {
322 const char* sep = strchr(benchmarks, ',');
323 Slice name;
324 if (sep == nullptr) {
325 name = benchmarks;
326 benchmarks = nullptr;
327 } else {
328 name = Slice(benchmarks, sep - benchmarks);
329 benchmarks = sep + 1;
330 }
331
332 Start();
333
334 bool known = true;
335 bool write_sync = false;
336 if (name == Slice("fillseq")) {
337 Write(write_sync, SEQUENTIAL, FRESH, num_, FLAGS_value_size, 1);
338 DBSynchronize(db_);
339 } else if (name == Slice("fillrandom")) {
340 Write(write_sync, RANDOM, FRESH, num_, FLAGS_value_size, 1);
341 DBSynchronize(db_);
342 } else if (name == Slice("overwrite")) {
343 Write(write_sync, RANDOM, EXISTING, num_, FLAGS_value_size, 1);
344 DBSynchronize(db_);
345 } else if (name == Slice("fillrandsync")) {
346 write_sync = true;
347 Write(write_sync, RANDOM, FRESH, num_ / 100, FLAGS_value_size, 1);
348 DBSynchronize(db_);
349 } else if (name == Slice("fillseqsync")) {
350 write_sync = true;
351 Write(write_sync, SEQUENTIAL, FRESH, num_ / 100, FLAGS_value_size, 1);
352 DBSynchronize(db_);
353 } else if (name == Slice("fillrand100K")) {
354 Write(write_sync, RANDOM, FRESH, num_ / 1000, 100 * 1000, 1);
355 DBSynchronize(db_);
356 } else if (name == Slice("fillseq100K")) {
357 Write(write_sync, SEQUENTIAL, FRESH, num_ / 1000, 100 * 1000, 1);
358 DBSynchronize(db_);
359 } else if (name == Slice("readseq")) {
360 ReadSequential();
361 } else if (name == Slice("readrandom")) {
362 ReadRandom();
363 } else if (name == Slice("readrand100K")) {
364 int n = reads_;
365 reads_ /= 1000;
366 ReadRandom();
367 reads_ = n;
368 } else if (name == Slice("readseq100K")) {
369 int n = reads_;
370 reads_ /= 1000;
371 ReadSequential();
372 reads_ = n;
373 } else {
374 known = false;
375 if (name != Slice()) { // No error message for empty name
376 fprintf(stderr, "unknown benchmark '%s'\n", name.ToString().c_str());
377 }
378 }
379 if (known) {
380 Stop(name);
381 }
382 }
383 }
384
385 private:
386 void Open(bool sync) {
387 assert(db_ == nullptr);
388
389 // Initialize db_
390 db_ = new kyotocabinet::TreeDB();
391 char file_name[100];
392 db_num_++;
393 std::string test_dir;
394 Env::Default()->GetTestDirectory(&test_dir);
395 snprintf(file_name, sizeof(file_name), "%s/dbbench_polyDB-%d.kct",
396 test_dir.c_str(), db_num_);
397
398 // Create tuning options and open the database
399 int open_options =
400 kyotocabinet::PolyDB::OWRITER | kyotocabinet::PolyDB::OCREATE;
401 int tune_options =
402 kyotocabinet::TreeDB::TSMALL | kyotocabinet::TreeDB::TLINEAR;
403 if (FLAGS_compression) {
404 tune_options |= kyotocabinet::TreeDB::TCOMPRESS;
405 db_->tune_compressor(&comp_);
406 }
407 db_->tune_options(tune_options);
408 db_->tune_page_cache(FLAGS_cache_size);
409 db_->tune_page(FLAGS_page_size);
410 db_->tune_map(256LL << 20);
411 if (sync) {
412 open_options |= kyotocabinet::PolyDB::OAUTOSYNC;
413 }
414 if (!db_->open(file_name, open_options)) {
415 fprintf(stderr, "open error: %s\n", db_->error().name());
416 }
417 }
418
419 void Write(bool sync, Order order, DBState state, int num_entries,
420 int value_size, int entries_per_batch) {
421 // Create new database if state == FRESH
422 if (state == FRESH) {
423 if (FLAGS_use_existing_db) {
424 message_ = "skipping (--use_existing_db is true)";
425 return;
426 }
427 delete db_;
428 db_ = nullptr;
429 Open(sync);
430 Start(); // Do not count time taken to destroy/open
431 }
432
433 if (num_entries != num_) {
434 char msg[100];
435 snprintf(msg, sizeof(msg), "(%d ops)", num_entries);
436 message_ = msg;
437 }
438
439 // Write to database
440 for (int i = 0; i < num_entries; i++) {
441 const int k = (order == SEQUENTIAL) ? i : (rand_.Next() % num_entries);
442 char key[100];
443 snprintf(key, sizeof(key), "%016d", k);
444 bytes_ += value_size + strlen(key);
445 std::string cpp_key = key;
446 if (!db_->set(cpp_key, gen_.Generate(value_size).ToString())) {
447 fprintf(stderr, "set error: %s\n", db_->error().name());
448 }
449 FinishedSingleOp();
450 }
451 }
452
453 void ReadSequential() {
454 kyotocabinet::DB::Cursor* cur = db_->cursor();
455 cur->jump();
456 std::string ckey, cvalue;
457 while (cur->get(&ckey, &cvalue, true)) {
458 bytes_ += ckey.size() + cvalue.size();
459 FinishedSingleOp();
460 }
461 delete cur;
462 }
463
464 void ReadRandom() {
465 std::string value;
466 for (int i = 0; i < reads_; i++) {
467 char key[100];
468 const int k = rand_.Next() % reads_;
469 snprintf(key, sizeof(key), "%016d", k);
470 db_->get(key, &value);
471 FinishedSingleOp();
472 }
473 }
474 };
475
476 } // namespace leveldb
477
478 int main(int argc, char** argv) {
479 std::string default_db_path;
480 for (int i = 1; i < argc; i++) {
481 double d;
482 int n;
483 char junk;
484 if (leveldb::Slice(argv[i]).starts_with("--benchmarks=")) {
485 FLAGS_benchmarks = argv[i] + strlen("--benchmarks=");
486 } else if (sscanf(argv[i], "--compression_ratio=%lf%c", &d, &junk) == 1) {
487 FLAGS_compression_ratio = d;
488 } else if (sscanf(argv[i], "--histogram=%d%c", &n, &junk) == 1 &&
489 (n == 0 || n == 1)) {
490 FLAGS_histogram = n;
491 } else if (sscanf(argv[i], "--num=%d%c", &n, &junk) == 1) {
492 FLAGS_num = n;
493 } else if (sscanf(argv[i], "--reads=%d%c", &n, &junk) == 1) {
494 FLAGS_reads = n;
495 } else if (sscanf(argv[i], "--value_size=%d%c", &n, &junk) == 1) {
496 FLAGS_value_size = n;
497 } else if (sscanf(argv[i], "--cache_size=%d%c", &n, &junk) == 1) {
498 FLAGS_cache_size = n;
499 } else if (sscanf(argv[i], "--page_size=%d%c", &n, &junk) == 1) {
500 FLAGS_page_size = n;
501 } else if (sscanf(argv[i], "--compression=%d%c", &n, &junk) == 1 &&
502 (n == 0 || n == 1)) {
503 FLAGS_compression = (n == 1) ? true : false;
504 } else if (strncmp(argv[i], "--db=", 5) == 0) {
505 FLAGS_db = argv[i] + 5;
506 } else {
507 fprintf(stderr, "Invalid flag '%s'\n", argv[i]);
508 exit(1);
509 }
510 }
511
512 // Choose a location for the test database if none given with --db=<path>
513 if (FLAGS_db == nullptr) {
514 leveldb::Env::Default()->GetTestDirectory(&default_db_path);
515 default_db_path += "/dbbench";
516 FLAGS_db = default_db_path.c_str();
517 }
518
519 leveldb::Benchmark benchmark;
520 benchmark.Run();
521 return 0;
522 }
523