Produced by Araxis Merge on 11/16/2017 4:02:38 PM Eastern Standard Time. See www.araxis.com for information about Merge. This report uses XHTML and CSS2, and is best viewed with a modern standards-compliant browser. For optimum results when printing this report, use landscape orientation and enable printing of background images and colours in your browser.
| # | Location | File | Last Modified |
|---|---|---|---|
| 1 | Thu Nov 16 21:02:37 2017 UTC | ||
| 2 | CUI-CPP-v2.3.1-release.zip\public\fake_docstore | 2016_rubykaigi.docx | Thu Nov 16 20:12:37 2017 UTC |
| Description | Between Files 1 and 2 |
|
|---|---|---|
| Text Blocks | Lines | |
| Unchanged | 0 | 0 |
| Changed | 0 | 0 |
| Inserted | 1 | 1199 |
| Removed | 0 | 0 |
| Whitespace | |
|---|---|
| Character case | Differences in character case are significant |
| Line endings | Differences in line endings (CR and LF characters) are ignored |
| CR/LF characters | Not shown in the comparison detail |
No regular expressions were active.
| 1 | A proposal of | |||||
| 2 | new concur rency mode l for Ruby 3 | |||||
| 3 | ||||||
| 4 | Koichi Sas ada | |||||
| 5 | ko1@heroku .com | |||||
| 6 | ||||||
| 7 | People lov e “Concurr ency” | |||||
| 8 | ||||||
| 9 | ||||||
| 10 | ||||||
| 11 | ||||||
| 12 | ||||||
| 13 | ||||||
| 14 | ||||||
| 15 | ||||||
| 16 | ||||||
| 17 | ||||||
| 18 | ||||||
| 19 | Concurrent | |||||
| 20 | RubyKaigi | |||||
| 21 | (at least, there are two | |||||
| 22 | parallel s essions) | |||||
| 23 | ||||||
| 24 | Why people love (to discuss) | |||||
| 25 | “Concurren cy”? | |||||
| 26 | • Performa nce by “Pa rallel” ex ecution to utilize | |||||
| 27 | multiple-c ores | |||||
| 28 | • Ruby has thread sy stem, but MRI doesn’ t permit t o allow pa rallel exe cution. | |||||
| 29 | ||||||
| 30 | About this presentat ion | |||||
| 31 | • Show “Wh y difficul t multi-th reads prog rams” | |||||
| 32 | • Propose new concur rent and p arallel me chanism id ea named “ Guild” | |||||
| 33 | For Ruby 3 | |||||
| 34 | ||||||
| 35 | Koichi Sas ada | |||||
| 36 | •A program mer living in Tokyo, Japan | |||||
| 37 | •Ruby core committer since 200 7 | |||||
| 38 | •YARV, Fib er, … (Rub y 1.9) | |||||
| 39 | •RGenGC, R incGC(Ruby 2…) | |||||
| 40 | ||||||
| 41 | Koichi is an Employe e | |||||
| 42 | ||||||
| 43 | ||||||
| 44 | ||||||
| 45 | ||||||
| 46 | ||||||
| 47 | ||||||
| 48 | ||||||
| 49 | ||||||
| 50 | ||||||
| 51 | ||||||
| 52 | ||||||
| 53 | ||||||
| 54 | ||||||
| 55 | ||||||
| 56 | ||||||
| 57 | ||||||
| 58 | ||||||
| 59 | ||||||
| 60 | Difficulty of | |||||
| 61 | Multi-thre ads progra mming | |||||
| 62 | ||||||
| 63 | Programmin g language evolution | |||||
| 64 | • Trade-of f: Perform ance v.s. Safety/Eas ily | |||||
| 65 | Performanc e: making faster pro grams | |||||
| 66 | ||||||
| 67 | Safety: ma king bug-f ree progra ms | |||||
| 68 | Easily: ma king progr ams with s mall effor ts | |||||
| 69 | ||||||
| 70 | Two exampl e | |||||
| 71 | C language | |||||
| 72 | • String m anipulatio n with poi nters | |||||
| 73 | • Memory m anagement without GC | |||||
| 74 | ||||||
| 75 | String man ipulation with point ers | |||||
| 76 | • C: Using raw point ers to man ipulate st rings | |||||
| 77 | Good: all- purpose an d fast | |||||
| 78 | Bad: Error -prone | |||||
| 79 | Generates strange be havior, su ch as abno rmal termi nation | |||||
| 80 | • Ruby: Wr ap with St ring class | |||||
| 81 | Good: Easy to use | |||||
| 82 | Bad: slowe r than C i n some cas es | |||||
| 83 | ||||||
| 84 | Object man agement wi thout GC | |||||
| 85 | • C: Free memory obj ects manua lly | |||||
| 86 | Good: full control ( target, ti ming and s o on) | |||||
| 87 | Bad: Error -prone | |||||
| 88 | double-fre e/memory-l eak, … | |||||
| 89 | • Ruby: Au tomatic co llection w ith GC | |||||
| 90 | Good: noth ing to car e about ob ject colle ction | |||||
| 91 | Bad: intro duce some overhead | |||||
| 92 | ||||||
| 93 | Ruby chose “safety/e asily” app roach | |||||
| 94 | •Ruby enco urage “Hap py Program ming” | |||||
| 95 | •Reduce pr ogrammer’s cost | |||||
| 96 | •Nowadays computer i s enough f aster | |||||
| 97 | •Implement ation tech niques ove rcome | |||||
| 98 | performanc e penaltie s | |||||
| 99 | ||||||
| 100 | Do you wan t to progr am without GC? | |||||
| 101 | ||||||
| 102 | Muilti-thr eads progr amming is difficult | |||||
| 103 | • Introduc e data rac e, race co ndition | |||||
| 104 | ||||||
| 105 | ||||||
| 106 | • Introduc e deadlock , livelock | |||||
| 107 | • Difficul ty on debu gging beca use of | |||||
| 108 | nondetermi nistic beh avior | |||||
| 109 | difficult to reprodu ce same pr oblem | |||||
| 110 | Difficult to make co rrect (bug -free) | |||||
| 111 | programs | |||||
| 112 | ||||||
| 113 | ||||||
| 114 | ||||||
| 115 | ||||||
| 116 | ||||||
| 117 | • Difficul t to tune performanc e | |||||
| 118 | Difficult to make | |||||
| 119 | fast progr ams | |||||
| 120 | ||||||
| 121 | Data race and race c ondition | |||||
| 122 | • Bank amo unt transf er example | |||||
| 123 | Quoted fro m Race Con dition vs. Data Race http://bl og.regehr. org/archiv es/490 | |||||
| 124 | ||||||
| 125 | ||||||
| 126 | ||||||
| 127 | def transf er1 (amoun t, account _from, acc ount_to) i f (account _from.bala nce < amou nt) return NOPE acco unt_to.bal ance += am ount accou nt_from.ba lance -= a mount | |||||
| 128 | return YEP | |||||
| 129 | end | |||||
| 130 | ||||||
| 131 | Data race | |||||
| 132 | •“account_ to.balance += amount ” has Data -race | |||||
| 133 | Assume two threads ( T1 and T2) invoke th is methods with same bank acco unts | |||||
| 134 | ||||||
| 135 | ||||||
| 136 | ||||||
| 137 | ||||||
| 138 | # interlea ve two thr eads (T1: amount = 1 00, T2: am ount = 200 ) T1: t1 = account_t o.balance # t1 = 10 | |||||
| 139 | T2: t2 = a ccount_to. balance # t2 = 10 | |||||
| 140 | T2: accoun t_to.balan ce = t2 + 200 #=> 21 0 | |||||
| 141 | T1: accoun t_to.balan ce = t1 + 100 #=> 11 0 (expecte d: 310) | |||||
| 142 | ||||||
| 143 | Race condi tion | |||||
| 144 | • To avoid data-race with the lock | |||||
| 145 | • But ther e is anoth er problem yet | |||||
| 146 | ||||||
| 147 | ||||||
| 148 | # Lock wit h “Thread. exclusive” | |||||
| 149 | def transf er2 (amoun t, account _from, acc ount_to) i f (account _from.bala nce < amou nt) return NOPE Thr ead.exclus ive{ accou nt_to.bala nce += amo unt } | |||||
| 150 | Thread.exc lusive{ ac count_from .balance - = amount } | |||||
| 151 | return YEP end | |||||
| 152 | ||||||
| 153 | Race condi tion | |||||
| 154 | • To avoid data-race with the lock | |||||
| 155 | • But ther e is anoth er problem yet | |||||
| 156 | ||||||
| 157 | ||||||
| 158 | ||||||
| 159 | # T1 amoun t = 100, T 2 amount = 200, acco unt_from.b alance = 2 50 | |||||
| 160 | T1: if (ac count_from .balance ( == 250) < 100) retur n NOPE # O K, go thro ugh | |||||
| 161 | T2: if (ac count_from .balance ( == 250) < 200) retur n NOPE T2: Thread.exc lusive{ ac count_to.b alance += 200 } | |||||
| 162 | T2:Thread. exclusive{ account_f rom.balanc e -= 200 } #=> 250-2 00 => 50 | |||||
| 163 | T1:Thread. exclusive{ account_t o.balance += 100 } | |||||
| 164 | T1:Thread. exclusive{ account_f rom.balanc e -= 100 } #=> 50 - 100 => neg ative numb er!! | |||||
| 165 | ||||||
| 166 | Final solu tion | |||||
| 167 | • Lock who le of meth od | |||||
| 168 | ||||||
| 169 | ||||||
| 170 | ||||||
| 171 | def transf er1 (amoun t, account _from, acc ount_to) | |||||
| 172 | Thread.exc lusive{ | |||||
| 173 | if (accoun t_from.bal ance < amo unt) retur n NOPE acc ount_to.ba lance += a mount acco unt_from.b alance -= amount | |||||
| 174 | return YEP | |||||
| 175 | } | |||||
| 176 | end | |||||
| 177 | ||||||
| 178 | Another ex ample | |||||
| 179 | Multi-thre ad quiz | |||||
| 180 | • What hap pen on thi s program? | |||||
| 181 | ||||||
| 182 | ||||||
| 183 | ||||||
| 184 | ary = [1, 2, 3] | |||||
| 185 | t1 = Threa d.new{ ary .concat [4 , 5, 6] | |||||
| 186 | } | |||||
| 187 | t2 = Threa d.new{ | |||||
| 188 | p ary # wh at ’s happ en? | |||||
| 189 | }.join | |||||
| 190 | ||||||
| 191 | (1) [1, 2, 3] | |||||
| 192 | (2) [1, 2, 3, 4, 5, 6] | |||||
| 193 | (3) (1) or (2) | |||||
| 194 | ||||||
| 195 | Another ex ample | |||||
| 196 | Multi-thre ad quiz | |||||
| 197 | • Answer: (4) depend s on an in terpreter | |||||
| 198 | ||||||
| 199 | ||||||
| 200 | ||||||
| 201 | ary = [1, 2, 3] | |||||
| 202 | t1 = Threa d.new{ ary .concat [4 , 5, 6] | |||||
| 203 | } | |||||
| 204 | t2 = Threa d.new{ | |||||
| 205 | p ary # wh at ’s happ en? | |||||
| 206 | }.join | |||||
| 207 | On MRI, (3 ) is corre ct | |||||
| 208 | ||||||
| 209 | It will sh ows | |||||
| 210 | [1, 2, 3] or | |||||
| 211 | [1, 2, 3, 4, 5, 6] | |||||
| 212 | (depends o n thread | |||||
| 213 | switching timing) | |||||
| 214 | ||||||
| 215 | Another ex ample | |||||
| 216 | Multi-thre ad quiz | |||||
| 217 | • Answer: (4) depend s on an in terpreter | |||||
| 218 | ||||||
| 219 | ||||||
| 220 | ||||||
| 221 | ary = [1, 2, 3] | |||||
| 222 | t1 = Threa d.new{ | |||||
| 223 | ary.concat [4, 5, 6] | |||||
| 224 | } | |||||
| 225 | t2 = Threa d.new{ | |||||
| 226 | p ary # wh at ’s happ en? | |||||
| 227 | }.join | |||||
| 228 | On JRuby: | |||||
| 229 | ||||||
| 230 | It can cau se Java ex ception be cause “Arr ay#concat” is not th read safe | |||||
| 231 | ||||||
| 232 | On JRuby … | |||||
| 233 | ||||||
| 234 | ||||||
| 235 | # similar program h = Hash.new (0) NA = 1 _000 | |||||
| 236 | 10_000.tim es{ ary = [] (1..10) .each{ Thr ead.new{ | |||||
| 237 | NA.times{| i| ary.con cat [i] | |||||
| 238 | } | |||||
| 239 | } | |||||
| 240 | } | |||||
| 241 | t2 = Threa d.new{ s = ary.dup | |||||
| 242 | }.join | |||||
| 243 | }Unhandled Java exce ption: jav a.lang.Nul lPointerEx ception | |||||
| 244 | ||||||
| 245 | java.lang. NullPointe rException : null | |||||
| 246 | rbInspect at org/jru by/RubyBas icObject.j ava:1105 i nspect at org/jruby/ RubyObject .java:516 | |||||
| 247 | inspectAry at org/jr uby/RubyAr ray.java:1 469 inspec t at org/j ruby/RubyA rray.java: 1497 | |||||
| 248 | cacheAndCa ll at org/ jruby/runt ime/callsi te/Caching CallSite.j ava:293 ca ll at org/ jruby/runt ime/callsi te/Caching CallSite.j ava:131 | |||||
| 249 | block in t .rb at t.r b:17 | |||||
| 250 | yieldDirec t at org/j ruby/runti me/Compile dIRBlockBo dy.java:15 6 yieldSpe cific at o rg/jruby/r untime/IRB lockBody.j ava:73 yie ldSpecific at org/jr uby/runtim e/Block.ja va:136 | |||||
| 251 | times at o rg/jruby/R ubyFixnum. java:291 | |||||
| 252 | cacheAndCa ll at org/ jruby/runt ime/callsi te/Caching CallSite.j ava:303 ca llBlock at org/jruby /runtime/c allsite/Ca chingCallS ite.java:1 41 | |||||
| 253 | call at or g/jruby/ru ntime/call site/Cachi ngCallSite .java:145 | |||||
| 254 | <top> at t .rb:3 | |||||
| 255 | invokeWith Arguments at java/la ng/invoke/ MethodHand le.java:59 9 load at org/jruby/ ir/Compile r.java:111 | |||||
| 256 | runScript at org/jru by/Ruby.ja va:833 run Script at org/jruby/ Ruby.java: 825 runNor mally at o rg/jruby/R uby.java:7 60 runFrom Main at or g/jruby/Ru by.java:57 9 doRunFro mMain at o rg/jruby/M ain.java:4 25 interna lRun at or g/jruby/Ma in.java:31 3 | |||||
| 257 | run at org /jruby/Mai n.java:242 main at o rg/jruby/M ain.java:2 04 | |||||
| 258 | ||||||
| 259 | jruby 9.1. 2.0 (2.3.0 ) 2016-05- 26 7357c8f OpenJDK 6 4-Bit Serv er VM 24.9 5-b01 on 1 .7.0_101-b 00 +jit [l inux-x86_6 4] | |||||
| 260 | On 8 hardw are thread s machine | |||||
| 261 | ||||||
| 262 | Difficulty of multi- threads pr ograms | |||||
| 263 | • We need to synchro nize all s haring mut able | |||||
| 264 | objects co rrectly | |||||
| 265 | We need to know whic h methods are thread -safe. | |||||
| 266 | Easy to tr ack all on small pro gram | |||||
| 267 | Difficult to track o n big prog rams, espe cially on | |||||
| 268 | programs u sing gems | |||||
| 269 | • We need to check a ll of sour ce codes, or believe | |||||
| 270 | library do cuments (b ut documen ts should be correct ) | |||||
| 271 | • Multi-th reads prog . requires “complete ness” | |||||
| 272 | ||||||
| 273 | Difficulty of multi- threads pr ograms (co nt.) | |||||
| 274 | • For debu gging, it is difficu lt to find out the b ugs | |||||
| 275 | Backtrace may not wo rk well be cause the problem ma y be place d on anoth er line. | |||||
| 276 | Bugs don’t appear fr equently w ith small data | |||||
| 277 | Difficult to reprodu ce issues because of | |||||
| 278 | nondetermi nistic beh avior | |||||
| 279 | ||||||
| 280 | FYI: | |||||
| 281 | Why MRI Ar ray#concat is thread -safe? | |||||
| 282 | • MRI uses GVL (Gian t/Global V M Lock) to control t hread swit ching timi ng and C m ethods (su ch as Arra y#concat) are workin g atomical ly. | |||||
| 283 | • GVL proh ibits para llel threa d executio n (BAD), h owever it avoids sev eral sever e issues ( GOOD). | |||||
| 284 | ||||||
| 285 | Thread pro gramming: | |||||
| 286 | Performanc e tuning i ssue | |||||
| 287 | ||||||
| 288 | ||||||
| 289 | a1 = []; a 2 = [] | |||||
| 290 | NA = 10_00 0_000 | |||||
| 291 | t1 = Threa d.new{ | |||||
| 292 | NA.times{| i| a1 << i } | |||||
| 293 | }.join | |||||
| 294 | t2 = Threa d.new{ | |||||
| 295 | NA.times{| i| a2 << i } | |||||
| 296 | }.join | |||||
| 297 | Serial pro gram: | |||||
| 298 | real | |||||
| 299 | user sys | |||||
| 300 | 0m8.568s | |||||
| 301 | 0m37.816s | |||||
| 302 | 0m5.530s | |||||
| 303 | on JRuby | |||||
| 304 | ||||||
| 305 | Thread pro gramming: | |||||
| 306 | Performanc e tuning i ssue | |||||
| 307 | ||||||
| 308 | ||||||
| 309 | a1 = []; a 2 = [] | |||||
| 310 | NA = 10_00 0_000 | |||||
| 311 | t1 = Threa d.new{ NA. times{|i| a1 << i } | |||||
| 312 | } | |||||
| 313 | t2 = Threa d.new{ NA. times{|i| a2 << i } | |||||
| 314 | } | |||||
| 315 | t1.join; t 2.join | |||||
| 316 | Parallel p rogram | |||||
| 317 | (2 threads ): | |||||
| 318 | real | |||||
| 319 | user | |||||
| 320 | sys | |||||
| 321 | 0m6.411s | |||||
| 322 | 0m20.527s | |||||
| 323 | 0m7.798s | |||||
| 324 | ||||||
| 325 | Thread pro gramming: | |||||
| 326 | Performanc e tuning i ssue | |||||
| 327 | ||||||
| 328 | ||||||
| 329 | a1 = []; a 2 = [] | |||||
| 330 | NA = 10_00 0_000 | |||||
| 331 | m1, m2= Mu tex.new, M utex.new t 1 = Thread .new{ | |||||
| 332 | NA.times{| i| m1.sync hronize{ a 1 << i }} | |||||
| 333 | } | |||||
| 334 | t2 = Threa d.new{ | |||||
| 335 | NA.times{| i| m2.sync hronize{ a 2 << i }} | |||||
| 336 | } | |||||
| 337 | t1.join; t 2.join | |||||
| 338 | Parallel p rogram wit h | |||||
| 339 | a useless lock 1 | |||||
| 340 | (2 threads ): | |||||
| 341 | real0m10.2 64s | |||||
| 342 | user | |||||
| 343 | sys | |||||
| 344 | 0m38.370s | |||||
| 345 | 0m4.406s | |||||
| 346 | ||||||
| 347 | Thread pro gramming: | |||||
| 348 | Performanc e tuning i ssue | |||||
| 349 | ||||||
| 350 | ||||||
| 351 | a1 = []; a 2 = [] | |||||
| 352 | NA = 10_00 0_000 | |||||
| 353 | m = Mutex. new t1 = T hread.new{ | |||||
| 354 | NA.times{| i| m.synch ronize{ a1 << i }} | |||||
| 355 | } | |||||
| 356 | t2 = Threa d.new{ | |||||
| 357 | NA.times{| i| m.synch ronize{ a2 << i }} | |||||
| 358 | } | |||||
| 359 | t1.join; t 2.join | |||||
| 360 | Parallel p rogram wit h | |||||
| 361 | a useless lock 2 | |||||
| 362 | (2 threads ): | |||||
| 363 | real0m15.1 63s | |||||
| 364 | user | |||||
| 365 | sys | |||||
| 366 | 0m45.317s | |||||
| 367 | 0m9.658s | |||||
| 368 | ||||||
| 369 | Performanc e tuning i ssue | |||||
| 370 | ||||||
| 371 | ||||||
| 372 | ||||||
| 373 | ||||||
| 374 | Execution time | |||||
| 375 | Serial pro gram | |||||
| 376 | 8.568s | |||||
| 377 | Parallel p rogram | |||||
| 378 | 6.411s | |||||
| 379 | Parallel p rogram wit h a | |||||
| 380 | useless lo ck 1 | |||||
| 381 | 10.264s | |||||
| 382 | Parallel p rogram wit h a useles s lock 2 | |||||
| 383 | 15.163s | |||||
| 384 | ||||||
| 385 | Thread pro gramming: | |||||
| 386 | Performanc e tuning i ssue | |||||
| 387 | ||||||
| 388 | We need to use just correct nu mber locks | |||||
| 389 | ||||||
| 390 | Not enough → unexpect ed behavio r | |||||
| 391 | Too much→ performanc e penalty | |||||
| 392 | ||||||
| 393 | FYI: synch ronization mechanism | |||||
| 394 | • Many syn chronizati on mechani sms… | |||||
| 395 | Mutual exc lusion (Mu tex), moni tor, criti cal sectio n | |||||
| 396 | Transactio nal memory (optimist ic lock) | |||||
| 397 | Atomic ins tructions | |||||
| 398 | Synchroniz ed Queue | |||||
| 399 | • … | |||||
| 400 | Research o n many lig htweight l ock algori thms | |||||
| 401 | • They ass ume we can use them correctly | |||||
| 402 | ||||||
| 403 | ||||||
| 404 | ||||||
| 405 | ||||||
| 406 | ||||||
| 407 | ||||||
| 408 | ||||||
| 409 | ||||||
| 410 | ||||||
| 411 | ||||||
| 412 | ||||||
| 413 | ||||||
| 414 | ||||||
| 415 | ||||||
| 416 | ||||||
| 417 | ||||||
| 418 | ||||||
| 419 | ||||||
| 420 | ||||||
| 421 | ||||||
| 422 | Overcome t hread diff iculty | |||||
| 423 | ||||||
| 424 | Key idea | |||||
| 425 | Problem: | |||||
| 426 | Easy to sh are mutabl e objects | |||||
| 427 | Idea: | |||||
| 428 | Do not all ow to shar e mutable objects | |||||
| 429 | without an y restrict ion | |||||
| 430 | ||||||
| 431 | Study from other lan guages | |||||
| 432 | • Shell sc ript with pipes, Rac ket (Place ) | |||||
| 433 | Copy mutab le data be tween proc esses w/ p ipes | |||||
| 434 | • Erlang/E lixir | |||||
| 435 | Do not all ow mutable data | |||||
| 436 | • Clojure | |||||
| 437 | Basically do not all ow mutable data | |||||
| 438 | Special da ta structu re to shar e mutable objects | |||||
| 439 | Note that it can sha re mutable objects o n Java lay er | |||||
| 440 | NOTE: we d o not list approache s using “t ype system ” | |||||
| 441 | ||||||
| 442 | ||||||
| 443 | ||||||
| 444 | ||||||
| 445 | ||||||
| 446 | ||||||
| 447 | ||||||
| 448 | ||||||
| 449 | Don’t you know Elixi r language ? | |||||
| 450 | ||||||
| 451 | ||||||
| 452 | ||||||
| 453 | ||||||
| 454 | ||||||
| 455 | ||||||
| 456 | ||||||
| 457 | ||||||
| 458 | Programmin g Elixir 1 .2 | |||||
| 459 | by Dave Th omas | |||||
| 460 | 邦訳：プログラミング Elixir | |||||
| 461 | 笹田耕一・鳥井雪共訳 2016/08/1 9 | |||||
| 462 | You can bu y it TODAY !! | |||||
| 463 | サイン会は明日13時 らしいです | |||||
| 464 | ||||||
| 465 | Summary of approache s | |||||
| 466 | Communicat ion with c opied data (shell sc ripts) | |||||
| 467 | Good: we d on’t need locks | |||||
| 468 | Bad: copy everything is slow | |||||
| 469 | Prohibit m utable obj ects | |||||
| 470 | Good: we d on’t need locks | |||||
| 471 | Bad: Ruby utilizes m any “write ” op erati ons . Unac ceptable. | |||||
| 472 | Provide sp ecial data structure to share mutable ob jects | |||||
| 473 | Good: we d on’t need locks (who don’t use such spec ial data | |||||
| 474 | structures ) | |||||
| 475 | Bad: Diffi cult to us e special data struc tures. | |||||
| 476 | ||||||
| 477 | ||||||
| 478 | ||||||
| 479 | ||||||
| 480 | ||||||
| 481 | ||||||
| 482 | ||||||
| 483 | ||||||
| 484 | ||||||
| 485 | ||||||
| 486 | ||||||
| 487 | ||||||
| 488 | ||||||
| 489 | ||||||
| 490 | ||||||
| 491 | ||||||
| 492 | ||||||
| 493 | ||||||
| 494 | ||||||
| 495 | ||||||
| 496 | Background wasfinish ed | |||||
| 497 | ||||||
| 498 | Our goal f or Ruby 3 | |||||
| 499 | • We need to keep co mpatibilit y with Rub y 2. | |||||
| 500 | • We can m ake parall el program . | |||||
| 501 | • We shoul dn’t consi der about locks any more. | |||||
| 502 | • We can s hare objec ts with co py, but co py operati on should be fast. | |||||
| 503 | • We shoul d share ob jects if w e can. | |||||
| 504 | • We can p rovide spe cial objec ts to shar e mutable objects li ke Clojure if we rea lly need s peed. | |||||
| 505 | ||||||
| 506 | ||||||
| 507 | ||||||
| 508 | ||||||
| 509 | ||||||
| 510 | ||||||
| 511 | ||||||
| 512 | ||||||
| 513 | ||||||
| 514 | “Guild” | |||||
| 515 | New concur rency mode l for Ruby 3 | |||||
| 516 | ||||||
| 517 | Guild: New concurren cy abstrac tion | |||||
| 518 | • Guild ha s at least one threa d (and a t hread has | |||||
| 519 | at least o ne fiber) | |||||
| 520 | ||||||
| 521 | ||||||
| 522 | Guild | |||||
| 523 | Thread | |||||
| 524 | Fiber | |||||
| 525 | Fiber | |||||
| 526 | Thread | |||||
| 527 | Fiber | |||||
| 528 | Guild | |||||
| 529 | ThreaGduil d FibeTrhr ead | |||||
| 530 | Fiber | |||||
| 531 | ||||||
| 532 | Threads in different guilds ca n run in | |||||
| 533 | Parallel | |||||
| 534 | Threads in different guilds ca n run in p arallel | |||||
| 535 | Threads in a same gu ild can no t run in p arallel | |||||
| 536 | because of GVL (or G GL: Giant Guild Lock ) | |||||
| 537 | ||||||
| 538 | ||||||
| 539 | ||||||
| 540 | G1:T1 G1:T 2 G2:T3 | |||||
| 541 | ||||||
| 542 | ||||||
| 543 | ||||||
| 544 | Acquire GG L | |||||
| 545 | Acquire GG L | |||||
| 546 | ||||||
| 547 | Guild and objects: | |||||
| 548 | All object s have the ir ownmemb ership | |||||
| 549 | • All of m utable obj ects shoul d belong t o only one | |||||
| 550 | Guild (all mutable o bjects are member of one guild ) | |||||
| 551 | • Other gu ilds can n ot access objects | |||||
| 552 | ||||||
| 553 | ||||||
| 554 | Guild 1 | |||||
| 555 | obj | |||||
| 556 | obj | |||||
| 557 | obj | |||||
| 558 | NG!! | |||||
| 559 | Can’t acce ss | |||||
| 560 | (read/writ e) | |||||
| 561 | Guild 2 | |||||
| 562 | obj | |||||
| 563 | obj | |||||
| 564 | ||||||
| 565 | Object mem bership | |||||
| 566 | Only one g uild can a ccess muta ble object | |||||
| 567 | → Wedon’t needtocons ideraboutl ocks | |||||
| 568 | ||||||
| 569 | Because: | |||||
| 570 | NO data ra ces and NO race cond itions | |||||
| 571 | (if all gu ilds use o nly one th read) | |||||
| 572 | ||||||
| 573 | Inter guil ds communi cation | |||||
| 574 | •“Guild::C hannel” to communica te each gu ilds | |||||
| 575 | •Two commu nication m ethods | |||||
| 576 | Copy | |||||
| 577 | Transfer m embership or Move in short | |||||
| 578 | ||||||
| 579 | Copy using Channel | |||||
| 580 | • Guild::C hannel#tra nsfer(obj) send deep copied | |||||
| 581 | object(s) to a desti nation gui ld. | |||||
| 582 | • dRuby an d multi-pr ocess syst em use thi s kind of communicat ion | |||||
| 583 | ||||||
| 584 | Copy using Channel | |||||
| 585 | ||||||
| 586 | ||||||
| 587 | channel.tr ansfer(o1) o1 = chann el.receive | |||||
| 588 | ||||||
| 589 | ||||||
| 590 | O2:Data | |||||
| 591 | O2:Data | |||||
| 592 | O3:Data | |||||
| 593 | O3:Data | |||||
| 594 | Guild1 | |||||
| 595 | Guild2 | |||||
| 596 | o1 | |||||
| 597 | channel | |||||
| 598 | o1 | |||||
| 599 | o2 | |||||
| 600 | o2 | |||||
| 601 | o3o3 | |||||
| 602 | COPY | |||||
| 603 | ||||||
| 604 | Move using Channel | |||||
| 605 | [New techn ique!!] | |||||
| 606 | Guild::Cha nnel#trans fer_member ship(obj) change | |||||
| 607 | the member ship of ob ject(s) | |||||
| 608 | Leave from the sourc e guild | |||||
| 609 | Join to th e destinat ion guild | |||||
| 610 | Prohibit a ccessing t o left obj ects | |||||
| 611 | Cause exce ptions and so on | |||||
| 612 | ex)obj = “ foo” | |||||
| 613 | ch.transfe r_membersh ip(obj) ob j.upcase#= > Error!! p(obj)#=> Error!! | |||||
| 614 | ||||||
| 615 | Move using Channel | |||||
| 616 | ||||||
| 617 | ||||||
| 618 | channel.tr ansfer_mem bership(o1 )o1 = chan nel.receiv e | |||||
| 619 | ||||||
| 620 | ||||||
| 621 | O2:Data | |||||
| 622 | O3:Data | |||||
| 623 | Guild1 | |||||
| 624 | Guild2 | |||||
| 625 | o1 | |||||
| 626 | channel | |||||
| 627 | o2 | |||||
| 628 | o3 | |||||
| 629 | MOVE | |||||
| 630 | ||||||
| 631 | Move using Channel | |||||
| 632 | ||||||
| 633 | ||||||
| 634 | channel.tr ansfer_mem bership(o1 )o1 = chan nel.receiv e | |||||
| 635 | ||||||
| 636 | ||||||
| 637 | O2:Data | |||||
| 638 | O3:Data | |||||
| 639 | From Guild 1 perspect ive, | |||||
| 640 | transferre d objects are invali dated | |||||
| 641 | Guild1 | |||||
| 642 | Guild2 | |||||
| 643 | - | |||||
| 644 | channel | |||||
| 645 | o1 | |||||
| 646 | - | |||||
| 647 | o2 | |||||
| 648 | -o3 | |||||
| 649 | MOVE | |||||
| 650 | ||||||
| 651 | Sharing im mutable ob jects | |||||
| 652 | • Immutabl e objects can be sha red with a ny | |||||
| 653 | guilds | |||||
| 654 | a1 = [1, 2 , 3].freez e: a1 is I mmutable o bject | |||||
| 655 | a2 = [1, O bject.new, 3].freeze : a2 is no t immutabl e | |||||
| 656 | • We only need to se nd referen ces | |||||
| 657 | very light weight, li ke thread- programmin g | |||||
| 658 | • Numeric objects, s ymbols, tr ue, false, nil are i mmutable(f rom Ruby 2 .0, 2.1, 2 .2) | |||||
| 659 | ||||||
| 660 | Sharing im mutable ob jects | |||||
| 661 | We can sha re referen ce to immu table obje cts | |||||
| 662 | channel.tr ansfer(o1) o1 = chann el.receive | |||||
| 663 | ||||||
| 664 | ||||||
| 665 | If o1 is i mmutable, any Guild can read o 1 | |||||
| 666 | O2:Data | |||||
| 667 | O3:Data | |||||
| 668 | Guild1 | |||||
| 669 | Ref to | |||||
| 670 | o1 | |||||
| 671 | read | |||||
| 672 | channel | |||||
| 673 | Guild2 | |||||
| 674 | Ref to o1 | |||||
| 675 | o1 | |||||
| 676 | read | |||||
| 677 | o2 | |||||
| 678 | o3 | |||||
| 679 | ||||||
| 680 | Use-case 1 : master – worker ty pe | |||||
| 681 | ||||||
| 682 | ||||||
| 683 | ||||||
| 684 | ||||||
| 685 | ||||||
| 686 | ||||||
| 687 | def fib(n) ... end | |||||
| 688 | g_fib = Gu ild.new(sc ript: %q{ ch = Guild .default_c hannel | |||||
| 689 | while n, r eturn_ch = ch.receiv e | |||||
| 690 | return_ch. transfer f ib(n) end | |||||
| 691 | }) | |||||
| 692 | ||||||
| 693 | ch = Guild ::Channel. new g_fib. transfer([ 3, ch]) | |||||
| 694 | p ch.recei ven, retur n_ch | |||||
| 695 | ||||||
| 696 | ||||||
| 697 | Main | |||||
| 698 | Guild | |||||
| 699 | ch | |||||
| 700 | Fibonacci | |||||
| 701 | Guild | |||||
| 702 | return_ch | |||||
| 703 | Answer of fib(n) | |||||
| 704 | NOTE: Maki ng other F ibonacci g uilds, | |||||
| 705 | you can co mpute fib( n) in para llel | |||||
| 706 | ||||||
| 707 | Use-case 2 : pipeline | |||||
| 708 | ||||||
| 709 | ||||||
| 710 | ||||||
| 711 | ||||||
| 712 | result_ch = Guild::C hannel.new g_pipe3 = Guild.new (script: % q{ | |||||
| 713 | while obj = Guild.de fault_chan nel.receiv e obj = mo dify_obj3( obj) Guild .argv[0].t ransfer_me mbership(o bj) end | |||||
| 714 | }, argv: [ result_ch] ) | |||||
| 715 | g_pipe2 = Guild.new( script: %q { | |||||
| 716 | while obj = Guild.de fault_chan nel.receiv e obj = mo dify_obj2( obj) Guild .argv[0].t ransfer_me mbership(o bj) end | |||||
| 717 | }, argv: [ g_pipe3]) | |||||
| 718 | g_pipe1 = Guild.new( script: %q { | |||||
| 719 | while obj = Guild.de fault_chan nel.receiv e obj = mo dify_obj1( obj) Guild .argv[0].t ransfer_me mbership(o bj) end | |||||
| 720 | }, argv: [ g_pipe2]) obj = Some Class.new | |||||
| 721 | g_pipe1.tr ansfer_mem bership(ob j) obj = r esult_ch.r eceiveMain | |||||
| 722 | Guild | |||||
| 723 | Obj’’’ | |||||
| 724 | ||||||
| 725 | ||||||
| 726 | Move | |||||
| 727 | Pipe 3 | |||||
| 728 | Guild | |||||
| 729 | ||||||
| 730 | ||||||
| 731 | ||||||
| 732 | Pipe 1 | |||||
| 733 | Guild | |||||
| 734 | obj | |||||
| 735 | ||||||
| 736 | ||||||
| 737 | ||||||
| 738 | Move | |||||
| 739 | and modify | |||||
| 740 | Obj’’’ | |||||
| 741 | Move | |||||
| 742 | and modify | |||||
| 743 | Pipe 2 | |||||
| 744 | Guild | |||||
| 745 | ||||||
| 746 | Obj’ | |||||
| 747 | ||||||
| 748 | Move | |||||
| 749 | and modify | |||||
| 750 | Obj’’ | |||||
| 751 | ||||||
| 752 | Use-case: | |||||
| 753 | Bank examp le | |||||
| 754 | ||||||
| 755 | ||||||
| 756 | ||||||
| 757 | Only bank guild main tains bank data | |||||
| 758 | ||||||
| 759 | ||||||
| 760 | ||||||
| 761 | ||||||
| 762 | g_bank = G uild.new(s cript: %q{ | |||||
| 763 | while acco unt_from, account_to , amount, ch = Guild .default_c hannel.rec eive | |||||
| 764 | if (Bank[a ccount_fro m].balance < amount) | |||||
| 765 | ch.transfe r :NOPE | |||||
| 766 | else | |||||
| 767 | Bank[accou nt_to].bal ance += am ount Bank[ account_fr om].balanc e -= amoun t ch.trans fer :YEP | |||||
| 768 | end end | |||||
| 769 | }) | |||||
| 770 | … | |||||
| 771 | Bank | |||||
| 772 | Guild | |||||
| 773 | requests | |||||
| 774 | Other | |||||
| 775 | guilds | |||||
| 776 | Other | |||||
| 777 | guilds | |||||
| 778 | ||||||
| 779 | Use-case: | |||||
| 780 | Introduce special da ta structu re | |||||
| 781 | ||||||
| 782 | ?? | |||||
| 783 | Other | |||||
| 784 | guilds | |||||
| 785 | Other | |||||
| 786 | guildsIdea s of speci al data st ructure to share mut able objec ts | |||||
| 787 | Use extern al RDB | |||||
| 788 | In process /external Key/value store | |||||
| 789 | Software t ransaction al | |||||
| 790 | memory | |||||
| 791 | • … | |||||
| 792 | ||||||
| 793 | Summary of use cases | |||||
| 794 | Making mul tiple work ers and co mpute in p arallel | |||||
| 795 | Requests a nd respons es are com municate v ia channel s | |||||
| 796 | You can se nd it with copy or m ove | |||||
| 797 | Maybe web applicatio n can empl oy this mo del | |||||
| 798 | Making Pip eline stru ctures and compute i n parallel | |||||
| 799 | Each task has own Gu ild | |||||
| 800 | Receive ta rget objec t, modify it and sen d it next pipeline | |||||
| 801 | You will s end it wit h move (tr ansfer mem bership) | |||||
| 802 | It will he lp applica tions like applying several fi lters for input data | |||||
| 803 | Own respon sibility b y one Guil d | |||||
| 804 | All access es are man aged by on e responsi ble Guild | |||||
| 805 | If you wan t to share mutable o bjects, we need spec ial data s tructures | |||||
| 806 | External R DBs or key /value sto res are al so good id ea for thi s purpose | |||||
| 807 | ||||||
| 808 | Communicat ion strate gy | |||||
| 809 | [Upper is better] | |||||
| 810 | • Passing immutable objects | |||||
| 811 | • Copy mut able objec ts | |||||
| 812 | • If you h ave perfor mance prob lem, move (transfer membership ) mutable objects | |||||
| 813 | • If you h ave perfor mance prob lem too, u se | |||||
| 814 | special da ta structu re to shar e mutable objects | |||||
| 815 | ||||||
| 816 | Compare be tween | |||||
| 817 | Thread mod el and Gui ld model | |||||
| 818 | • On threa ds, it is difficult to find ou t which ob jects | |||||
| 819 | are shared mutable o bjects | |||||
| 820 | • On Guild s, there a re no shar ed mutable objects | |||||
| 821 | If there a re special data stru cture to s hare mutab le | |||||
| 822 | objects, w e only nee d to check around th is code | |||||
| 823 | ||||||
| 824 | → Encourag e “Safe” a nd “Easy” programmin g | |||||
| 825 | ||||||
| 826 | Compare be tween | |||||
| 827 | Thread mod el and Gui ld model | |||||
| 828 | On threads , inter th reads comm unication is very fa st. | |||||
| 829 | On guilds, inter gui lds commun ication in troduce ov erhead | |||||
| 830 | “Move” (tr ansfer mem bership) t echnique c an reduce | |||||
| 831 | this kind of overhea ds | |||||
| 832 | ||||||
| 833 | Trade-off: Performan ce v.s. Sa fety/Easil y | |||||
| 834 | Which do y ou want to choose? | |||||
| 835 | ||||||
| 836 | Digression : The name of “Guild ” | |||||
| 837 | •“Guild” i s good met aphor for “object’s | |||||
| 838 | membership ” | |||||
| 839 | • Check du plication | |||||
| 840 | First lett er is not same as ot her simila r abstract ions | |||||
| 841 | For variab le names | |||||
| 842 | P is for P rocesses, T is for T hreads, F is for Fib ers | |||||
| 843 | There are no duplica ting top-l evel class es and | |||||
| 844 | modules in all of ru bygems | |||||
| 845 | ||||||
| 846 | Implementa tion of “G uild” | |||||
| 847 | • How to i mplement i nter Guild s communic ation | |||||
| 848 | • How to i solate pro cess globa l data | |||||
| 849 | ||||||
| 850 | Howto impl ement inte r Guilds | |||||
| 851 | communicat ion | |||||
| 852 | • Copy | |||||
| 853 | • Move (tr ansfer mem bership) | |||||
| 854 | ||||||
| 855 | Copy using Channel | |||||
| 856 | ||||||
| 857 | ||||||
| 858 | channel.tr ansfer(o1) o1 = chann el.receive | |||||
| 859 | ||||||
| 860 | ||||||
| 861 | O2:Data | |||||
| 862 | O2:Data | |||||
| 863 | O3:Data | |||||
| 864 | O3:Data | |||||
| 865 | Guild1 | |||||
| 866 | Guild2 | |||||
| 867 | o1 | |||||
| 868 | channel | |||||
| 869 | o1 | |||||
| 870 | o2 | |||||
| 871 | o2 | |||||
| 872 | o3o3 | |||||
| 873 | COPY | |||||
| 874 | ||||||
| 875 | Copy using Channel | |||||
| 876 | Implementa tion | |||||
| 877 | channel.tr ansfer(o1) o1 = chann el.receive | |||||
| 878 | ||||||
| 879 | ||||||
| 880 | O2:Data | |||||
| 881 | O3:Data | |||||
| 882 | Guild1 | |||||
| 883 | (1) Make | |||||
| 884 | deep copy | |||||
| 885 | Guild2 | |||||
| 886 | o1o1 | |||||
| 887 | channel | |||||
| 888 | o2 | |||||
| 889 | o2 | |||||
| 890 | o3o3 | |||||
| 891 | O2:Data | |||||
| 892 | O3:Data | |||||
| 893 | ||||||
| 894 | ||||||
| 895 | ||||||
| 896 | Copy using Channel I mplementat ion | |||||
| 897 | We can use CoW | |||||
| 898 | technique for data | |||||
| 899 | ||||||
| 900 | ||||||
| 901 | channel.tr ansfer(o1) o1 = chann el.receive | |||||
| 902 | ||||||
| 903 | ||||||
| 904 | O2:Data | |||||
| 905 | O2:Data | |||||
| 906 | O3:Data | |||||
| 907 | Guild1 | |||||
| 908 | Guild2 | |||||
| 909 | o1 | |||||
| 910 | channel | |||||
| 911 | o1 | |||||
| 912 | o2 | |||||
| 913 | o2 | |||||
| 914 | o3o3 | |||||
| 915 | O3:Data | |||||
| 916 | (2) Move/J oin | |||||
| 917 | ||||||
| 918 | Move using Channel | |||||
| 919 | ||||||
| 920 | ||||||
| 921 | channel.tr ansfer_mem bership(o1 )o1 = chan nel.receiv e | |||||
| 922 | ||||||
| 923 | ||||||
| 924 | O2:Data | |||||
| 925 | O3:Data | |||||
| 926 | Guild1 | |||||
| 927 | Guild2 | |||||
| 928 | o1 | |||||
| 929 | channel | |||||
| 930 | o2 | |||||
| 931 | o3 | |||||
| 932 | MOVE | |||||
| 933 | ||||||
| 934 | Move using Channel | |||||
| 935 | ||||||
| 936 | ||||||
| 937 | channel.tr ansfer_mem bership(o1 )o1 = chan nel.receiv e | |||||
| 938 | ||||||
| 939 | ||||||
| 940 | O2:Data | |||||
| 941 | O3:Data | |||||
| 942 | From Guild 1 perspect ive, | |||||
| 943 | transferre d objects are invali dated | |||||
| 944 | Guild1 | |||||
| 945 | Guild2 | |||||
| 946 | - | |||||
| 947 | channel | |||||
| 948 | o1 | |||||
| 949 | - | |||||
| 950 | o2 | |||||
| 951 | -o3 | |||||
| 952 | MOVE | |||||
| 953 | ||||||
| 954 | Move using Channel | |||||
| 955 | Implementa tion | |||||
| 956 | channel.tr ansfer_mem bership(o1 )o1 = chan nel.receiv e | |||||
| 957 | ||||||
| 958 | ||||||
| 959 | O2:Data | |||||
| 960 | O3:Data | |||||
| 961 | Guild1 | |||||
| 962 | (1) Make | |||||
| 963 | deep copy | |||||
| 964 | Guild2 | |||||
| 965 | o-1 | |||||
| 966 | (2) Invali date origi nals | |||||
| 967 | o-2 | |||||
| 968 | o-3 | |||||
| 969 | o1 | |||||
| 970 | channel | |||||
| 971 | o2 | |||||
| 972 | o3 | |||||
| 973 | ||||||
| 974 | Move using Channel | |||||
| 975 | Implementa tion | |||||
| 976 | channel.tr ansfer_mem bership(o1 )o1 = chan nel.receiv e | |||||
| 977 | ||||||
| 978 | ||||||
| 979 | O2:Data | |||||
| 980 | Guild1 | |||||
| 981 | Guild2 | |||||
| 982 | o-1 | |||||
| 983 | (2) Invali date origi nals | |||||
| 984 | o-2 | |||||
| 985 | o-3 | |||||
| 986 | channel | |||||
| 987 | o1 | |||||
| 988 | o2 | |||||
| 989 | o3 | |||||
| 990 | (3) Move/J oin | |||||
| 991 | O3:Data | |||||
| 992 | ||||||
| 993 | Ruby globa l data | |||||
| 994 | Global var iables ($f oo) | |||||
| 995 | Change the m to Guild local var iables | |||||
| 996 | Class and module obj ects | |||||
| 997 | Share betw een guilds | |||||
| 998 | Class vari ables | |||||
| 999 | Change the m to guild local. So that it i s guild/cl ass local variables | |||||
| 1000 | Constants | |||||
| 1001 | Share betw een guilds | |||||
| 1002 | However if assigned object is not a immu table obje ct, this c onstant is accessed only by se tting guil ds. If oth er guilds try to acc ess it, th em cause e rror. | |||||
| 1003 | Instance v ariables o f class an d module o bjects | |||||
| 1004 | Difficult. There are several a pproaches. | |||||
| 1005 | Proc/Bindi ng objects | |||||
| 1006 | Make it co py-able wi th env obj ects or en v independ ent object s | |||||
| 1007 | ObjectSpac e.each_obj ect | |||||
| 1008 | OMG | |||||
| 1009 | ||||||
| 1010 | Interprete r process global dat a | |||||
| 1011 | GC/Heap | |||||
| 1012 | Share it. Do stop th e world pa rallel mar king- and lazy concu rrent swee ping. | |||||
| 1013 | Synchroniz e only at page acqui re timing. No any sy nchronizat ion at cre ation time . | |||||
| 1014 | Inline met hod cache | |||||
| 1015 | To fill ne w entry, c reate an i nline cach e object a nd update atomically . | |||||
| 1016 | Tables (su ch as meth od tables and consta nt tables) | |||||
| 1017 | Introduce mutual exc lusions. | |||||
| 1018 | Current wo rking dire ctory (cwd ) | |||||
| 1019 | Each guild should ha ve own cwd (using op enat and s o on). | |||||
| 1020 | Signal | |||||
| 1021 | Design new signal de livery pro tocol and mechanism | |||||
| 1022 | C level gl obal varia bles | |||||
| 1023 | Avoid them . | |||||
| 1024 | Main guild can use C extension s depends on them | |||||
| 1025 | Current th read | |||||
| 1026 | Use TLS (t emporary), but we wi ll change all of C A PIs to rec eive conte xt data as first par ameter in the future . | |||||
| 1027 | ||||||
| 1028 | Performanc e evaluati on | |||||
| 1029 | • On 2 cor e virtual machine | |||||
| 1030 | Linux on V irtualBox on Windows 7 | |||||
| 1031 | • Now, we can’t run Ruby progr am on othe r than mai n guild, s o other gu ilds are i mplemented by C code | |||||
| 1032 | ||||||
| 1033 | Performanc e evaluati on | |||||
| 1034 | Simple num eric task in paralle l | |||||
| 1035 | ||||||
| 1036 | ||||||
| 1037 | Execution time (sec) | |||||
| 1038 | Single-Gui ld | |||||
| 1039 | 19.45 | |||||
| 1040 | Multi-Guil d | |||||
| 1041 | 10.45 | |||||
| 1042 | Fibonacci | |||||
| 1043 | ||||||
| 1044 | Main | |||||
| 1045 | Guild | |||||
| 1046 | FGibuoilnd acci | |||||
| 1047 | FGibuoilnd acci | |||||
| 1048 | GFiubioldn acci Guild | |||||
| 1049 | ||||||
| 1050 | ||||||
| 1051 | Total 50 r equests to compute f ib(40) Sen d 40 (inte ger) in ea ch request | |||||
| 1052 | ||||||
| 1053 | ||||||
| 1054 | ||||||
| 1055 | Execution | |||||
| 1056 | time (sec) | |||||
| 1057 | Single-Gui ld | |||||
| 1058 | 1.00 | |||||
| 1059 | Multi/ref | |||||
| 1060 | 0.64 | |||||
| 1061 | Multi/move | |||||
| 1062 | 4.29 | |||||
| 1063 | Multi/copy | |||||
| 1064 | 5.16 | |||||
| 1065 | Performanc e evaluati on | |||||
| 1066 | ||||||
| 1067 | Main | |||||
| 1068 | Guild | |||||
| 1069 | sum | |||||
| 1070 | sum | |||||
| 1071 | sum | |||||
| 1072 | sum GuildC opy/Move | |||||
| 1073 | ||||||
| 1074 | ||||||
| 1075 | ||||||
| 1076 | ||||||
| 1077 | ||||||
| 1078 | ||||||
| 1079 | ||||||
| 1080 | ||||||
| 1081 | ||||||
| 1082 | ||||||
| 1083 | ||||||
| 1084 | ||||||
| 1085 | ||||||
| 1086 | ||||||
| 1087 | ||||||
| 1088 | ||||||
| 1089 | ||||||
| 1090 | ||||||
| 1091 | ||||||
| 1092 | ||||||
| 1093 | ||||||
| 1094 | ||||||
| 1095 | ||||||
| 1096 | Total 100 requests t o compute sum of arr ay Send (1 ..10_000_0 00).to_a i n each req uest | |||||
| 1097 | Too slow!! | |||||
| 1098 | Because “m ove” need to | |||||
| 1099 | check all of element s | |||||
| 1100 | ||||||
| 1101 | Performanc e evaluati on | |||||
| 1102 | Copy/Move | |||||
| 1103 | ||||||
| 1104 | ||||||
| 1105 | Main | |||||
| 1106 | Guild | |||||
| 1107 | sum | |||||
| 1108 | sum | |||||
| 1109 | sum | |||||
| 1110 | sum Guild | |||||
| 1111 | ||||||
| 1112 | Execution time (sec) | |||||
| 1113 | Single-Gui ld | |||||
| 1114 | 1.00 | |||||
| 1115 | Multi/ref | |||||
| 1116 | 0.64 | |||||
| 1117 | Multi/move | |||||
| 1118 | 0.64 | |||||
| 1119 | ||||||
| 1120 | ||||||
| 1121 | If we know this arra y only has immutable objects, | |||||
| 1122 | we don’t n eed to che ck all ele ments => s pecial dat a structur e | |||||
| 1123 | ||||||
| 1124 | Check our goal for R uby 3 | |||||
| 1125 | We need to keep comp atibility with Ruby 2. | |||||
| 1126 | OK: Only i n main gui ld, it is compatible . | |||||
| 1127 | We can mak e parallel program. | |||||
| 1128 | OK: Guilds can run i n parallel . | |||||
| 1129 | We shouldn ’t conside r about lo cks any mo re. | |||||
| 1130 | OK: Only u sing copy and move, we don’t n eed to car e locks. | |||||
| 1131 | We can sha re objects with copy , but copy operation should be fast. | |||||
| 1132 | OK: Move ( transfer m embership) idea can reduce ove rhead. | |||||
| 1133 | We should share obje cts if we can. | |||||
| 1134 | OK: We can share imm utable obj ects fast and easily . | |||||
| 1135 | We can pro vide speci al objects to share mutable ob jects like Clojure i f we reall y need spe ed. | |||||
| 1136 | OK: Yes, w e can prov ide. | |||||
| 1137 | ||||||
| 1138 | Summary | |||||
| 1139 | • Introduc e “why thr eads are v ery diffic ult” | |||||
| 1140 | • Propose new concur rency abst raction “G uild” for Ruby 3 | |||||
| 1141 | Not implem ented ever ything yet , but I sh ow key ide as and pre liminary e valuation | |||||
| 1142 | ||||||
| 1143 | ||||||
| 1144 | ||||||
| 1145 | ||||||
| 1146 | Thank you for your a ttention | |||||
| 1147 | ||||||
| 1148 | Koichi Sas ada | |||||
| 1149 | <ko1@herok u.com> | |||||
| 1150 | ||||||
| 1151 | ||||||
| 1152 | ||||||
| 1153 | ||||||
| 1154 | ||||||
| 1155 | Approach c omparison | |||||
| 1156 | ||||||
| 1157 | ||||||
| 1158 | ||||||
| 1159 | ||||||
| 1160 | Process/MV M | |||||
| 1161 | Place (Rac ket) | |||||
| 1162 | Guild (cop y/move) | |||||
| 1163 | Thread | |||||
| 1164 | Heap | |||||
| 1165 | Separate | |||||
| 1166 | Separate | |||||
| 1167 | Share | |||||
| 1168 | Share | |||||
| 1169 | Communicat ion | |||||
| 1170 | Mutable ob jects | |||||
| 1171 | Copy | |||||
| 1172 | Copy | |||||
| 1173 | Copy/Move | |||||
| 1174 | Share | |||||
| 1175 | Communicat ion Frozen object | |||||
| 1176 | Copy | |||||
| 1177 | Share (may be) | |||||
| 1178 | Share | |||||
| 1179 | Share | |||||
| 1180 | Lock | |||||
| 1181 | Don’t need | |||||
| 1182 | Don’t need | |||||
| 1183 | (mostly) D on’t need | |||||
| 1184 | Required | |||||
| 1185 | ISeq | |||||
| 1186 | Copy | |||||
| 1187 | Share | |||||
| 1188 | Share | |||||
| 1189 | Share | |||||
| 1190 | Class/Modu le | |||||
| 1191 | Copy | |||||
| 1192 | Copy (fork ) | |||||
| 1193 | Share | |||||
| 1194 | Share | |||||
| 1195 | ||||||
| 1196 | ||||||
| 1197 | ||||||
| 1198 | ||||||
| 1199 |
Araxis Merge (but not the data content of this report) is Copyright © 1993-2016 Araxis Ltd (www.araxis.com). All rights reserved.