在內存分配期間使用多線程的Java

Question

我目前正在編寫一個java程序，用於分配一個大的多維數組進行處理。在分配過程中，java最終使用了所有可用的CPU（在我的情況下是8個CPU）。這是預期的行爲？我正在運行Java版本爲1.7.0_65的Ubuntu 14.04。如果您需要更多信息，請告訴我。如果這是預期的行爲，是否有辦法限制它，以便它只使用一個CPU？

public class test { 
    public static void main(String[] args){ 
     System.out.println("Test"); 

     int a = 1000; 
     int b = 100; 
     int c = 100; 
     int d = 1; 
     int e = 10; 
     float[][][][][] test = new float[a][b][c][d][e]; 


     System.out.println("Done"); 
    } 
} 

編輯：通過設置足夠大的初始堆大小，程序永遠不會結束使用多個CPU。在未來，我將使用足夠大的初始JVM大小和taskset命令來限制cpu affiinty。

Answer 1

總是在當前線程中執行分配，但是如果堆中沒有足夠的空間，則分配請求可以觸發垃圾回收，根據JVM及其垃圾回收器調整參數，可以使用多個線程。

在Oracle JVM上，可以通過向Java進程提供-verbose：gc選項來驗證確實如此。在我的電腦，你的程序，然後將產生以下輸出：

Test 
[GC (Allocation Failure) 512K->456K(261632K), 0.0007027 secs] 
[GC (Allocation Failure) 968K->920K(261632K), 0.0014950 secs] 
[GC (Allocation Failure) 1432K->1452K(261632K), 0.0011264 secs] 
[GC (Allocation Failure) 1964K->1972K(262144K), 0.0009485 secs] 
[GC (Allocation Failure) 2996K->3020K(262144K), 0.0011456 secs] 
[GC (Allocation Failure) 4044K->4172K(264704K), 0.0014029 secs] 
[GC (Allocation Failure) 6220K->6124K(264704K), 0.0032563 secs] 
[GC (Allocation Failure) 8172K->8212K(267776K), 0.0032141 secs] 
[GC (Allocation Failure) 11796K->11836K(268288K), 0.0048205 secs] 
[GC (Allocation Failure) 15420K->15524K(271872K), 0.0048205 secs] 
[GC (Allocation Failure) 20644K->20772K(272896K), 0.0063642 secs] 
[GC (Allocation Failure) 25892K->25900K(275968K), 0.0066470 secs] 
[GC (Allocation Failure) 33580K->33636K(276480K), 0.0069722 secs] 
[GC (Allocation Failure) 41316K->41396K(281088K), 0.0105088 secs] 
[GC (Allocation Failure) 52660K->52564K(282112K), 0.0095962 secs] 
[GC (Allocation Failure) 63828K->64060K(293376K), 0.0127885 secs] 
[GC (Allocation Failure) 81468K->81380K(294400K), 0.0138240 secs] 
[GC (Allocation Failure) 98788K->99004K(307712K), 0.0192218 secs] 
[GC (Allocation Failure) 122044K->122196K(309248K), 0.0170214 secs] 
[GC (Allocation Failure) 145236K->145284K(318976K), 0.0242662 secs] 
[GC (Allocation Failure) 174468K->174620K(322048K), 0.0219776 secs] 
[GC (Allocation Failure) 203804K->203860K(334336K), 0.0287066 secs] 
[GC (Allocation Failure) 241236K->241356K(337920K), 0.0252992 secs] 
[GC (Allocation Failure) 278732K->278916K(354816K), 0.0359923 secs] 
[Full GC (Ergonomics) 278916K->277284K(400384K), 1.5867443 secs] 
[GC (Allocation Failure) 325412K->325695K(404992K), 0.0360435 secs] 
[Full GC (Ergonomics) 325695K->325411K(449024K), 0.7776627 secs] 
[GC (Allocation Failure) 373539K->373719K(449024K), 0.0383219 secs] 
[Full GC (Ergonomics) 373719K->373537K(493056K), 0.8760845 secs] 
[GC (Allocation Failure) 414497K->414792K(506368K), 0.0410957 secs] 
[Full GC (Ergonomics) 414792K->414495K(553472K), 0.9881370 secs] 
[GC (Allocation Failure) 455455K->455705K(549376K), 0.0421990 secs] 
[GC (Allocation Failure) 491545K->491545K(557568K), 0.0329242 secs] 
[Full GC (Ergonomics) 491545K->491290K(604160K), 1.1687245 secs] 
[GC (Allocation Failure) 527130K->527471K(621568K), 0.0546765 secs] 
[Full GC (Ergonomics) 527471K->527127K(676864K), 1.2818099 secs] 
[GC (Allocation Failure) 565527K->565750K(679424K), 0.0519078 secs] 
[GC (Allocation Failure) 604150K->604174K(669184K), 0.0404352 secs] 
[GC (Allocation Failure) 623630K->623910K(690176K), 0.0469939 secs] 
[Full GC (Ergonomics) 623910K->623379K(753152K), 1.6304653 secs] 
[GC (Allocation Failure) 642835K->643153K(751616K), 0.0642752 secs] 
[GC (Allocation Failure) 655441K->655473K(760832K), 0.0408013 secs] 
[GC (Allocation Failure) 667761K->667761K(760320K), 0.0448666 secs] 
[GC (Allocation Failure) 668273K->668273K(773632K), 0.0452070 secs] 
[GC (Allocation Failure) 668785K->668785K(784896K), 0.0464397 secs] 
[GC (Allocation Failure) 669297K->669297K(785408K), 0.0455053 secs] 
[GC (Allocation Failure) 669809K->669841K(799232K), 0.0461530 secs] 
[GC (Allocation Failure) 670353K->670401K(696320K), 0.0496845 secs] 
[Full GC (Ergonomics) 670401K->669956K(762880K), 1.6411686 secs] 
[GC (Allocation Failure) 670468K->670818K(855552K), 0.0602765 secs] 
[GC (Allocation Failure) 671842K->671778K(721408K), 0.0332749 secs] 
[GC (Allocation Failure) 672802K->672770K(846848K), 0.0333530 secs] 
[GC (Allocation Failure) 674306K->674426K(711680K), 0.0328486 secs] 
[Full GC (Ergonomics) 674426K->674045K(778240K), 0.6690573 secs] 
[GC (Allocation Failure) 675581K->675704K(913408K), 0.0583846 secs] 
[GC (Allocation Failure) 676216K->676184K(914944K), 0.0275571 secs] 
[GC (Allocation Failure) 676696K->676760K(912896K), 0.0276851 secs] 
[GC (Allocation Failure) 677784K->677768K(744448K), 0.0277325 secs] 
[GC (Allocation Failure) 678792K->678824K(903168K), 0.0281344 secs] 
[GC (Allocation Failure) 680360K->680360K(747520K), 0.0280448 secs] 
[GC (Allocation Failure) 681896K->681896K(889856K), 0.0294963 secs] 
[GC (Allocation Failure) 683944K->683912K(751616K), 0.0298778 secs] 
[GC (Allocation Failure) 685960K->685992K(874496K), 0.0305114 secs] 
[GC (Allocation Failure) 688552K->688616K(756736K), 0.0308301 secs] 
[GC (Allocation Failure) 691176K->691144K(860160K), 0.0315098 secs] 
[GC (Allocation Failure) 694728K->694696K(763904K), 0.0317261 secs] 
[GC (Allocation Failure) 698280K->698312K(844800K), 0.0324186 secs] 
[GC (Allocation Failure) 702920K->702952K(773120K), 0.0329357 secs] 
[GC (Allocation Failure) 707560K->707616K(834048K), 0.0336230 secs] 
[GC (Allocation Failure) 713248K->713264K(782336K), 0.0331354 secs] 
[GC (Allocation Failure) 718896K->718928K(835072K), 0.0336960 secs] 
[GC (Allocation Failure) 726096K->726200K(836608K), 0.0341606 secs] 
[GC (Allocation Failure) 733368K->733344K(836608K), 0.0368486 secs] 
[GC (Allocation Failure) 740000K->740064K(838656K), 0.0377613 secs] 
[GC (Allocation Failure) 746720K->746768K(839680K), 0.0393101 secs] 
[GC (Allocation Failure) 750864K->750864K(843776K), 0.0390566 secs] 
[GC (Allocation Failure) 754960K->755064K(845312K), 0.0392218 secs] 
[GC (Allocation Failure) 756600K->756640K(848896K), 0.0383347 secs] 
[GC (Allocation Failure) 758176K->758208K(850432K), 0.0379328 secs] 
[GC (Allocation Failure) 759232K->759200K(804864K), 0.0378035 secs] 
[GC (Allocation Failure) 760224K->760304K(846848K), 0.0400614 secs] 
[GC (Allocation Failure) 761840K->761848K(795136K), 0.0382733 secs] 
[GC (Allocation Failure) 763384K->763432K(835584K), 0.0364710 secs] 
[GC (Allocation Failure) 765480K->765568K(787968K), 0.0364749 secs] 
[GC (Allocation Failure) 767616K->767632K(834048K), 0.0357376 secs] 
[Full GC (Ergonomics) 767632K->766656K(881152K), 1.1594624 secs] 
[GC (Allocation Failure) 769216K->769444K(882688K), 0.0689485 secs] 
[GC (Allocation Failure) 772004K->772044K(884224K), 0.0334374 secs] 
[GC (Allocation Failure) 775116K->775188K(884224K), 0.0350246 secs] 
[GC (Allocation Failure) 778260K->778300K(881664K), 0.0336294 secs] 
[GC (Allocation Failure) 782396K->782396K(810496K), 0.0337357 secs] 
[GC (Allocation Failure) 786492K->786524K(877056K), 0.0349286 secs] 
[GC (Allocation Failure) 791644K->791684K(816640K), 0.0353728 secs] 
[GC (Allocation Failure) 796804K->796876K(869888K), 0.0359552 secs] 
[GC (Allocation Failure) 803020K->803068K(820736K), 0.0366515 secs] 
[Full GC (Ergonomics) 803068K->802478K(869888K), 1.8098995 secs] 
[GC (Allocation Failure) 808622K->808797K(913920K), 0.0726835 secs] 
[GC (Allocation Failure) 815965K->815997K(872448K), 0.0363674 secs] 
Done 

正如你所看到的，JVM花費大量的時間與垃圾收集反覆嘗試。這是因爲在增加堆大小之前，JVM將首先嚐試收集垃圾。您可以通過指示JVM從適當的大堆開始來避免這種情況。再次假設一個Oracle JVM，您可以通過在命令行上提供-Xms1g來完成此操作。然後，產量遠遠短：

Test 
[GC (Allocation Failure) 511K->496K(1048064K), 0.0007194 secs] 
[GC (Allocation Failure) 1008K->904K(1048064K), 0.0014771 secs] 
[GC (Allocation Failure) 1416K->1456K(1048064K), 0.0011328 secs] 
[GC (Allocation Failure) 1968K->1996K(1048576K), 0.0011072 secs] 
[GC (Allocation Failure) 3020K->3068K(1048576K), 0.0011904 secs] 
[GC (Allocation Failure) 4092K->4252K(1051136K), 0.0012557 secs] 
[GC (Allocation Failure) 6300K->6148K(1051136K), 0.0033741 secs] 
[GC (Allocation Failure) 8196K->8228K(1054208K), 0.0032742 secs] 
[GC (Allocation Failure) 11812K->11852K(1054720K), 0.0049498 secs] 
[GC (Allocation Failure) 15436K->15604K(1058304K), 0.0042598 secs] 
[GC (Allocation Failure) 20724K->20772K(1059328K), 0.0058598 secs] 
[GC (Allocation Failure) 25892K->25972K(1062400K), 0.0061824 secs] 
[GC (Allocation Failure) 33652K->33700K(1062912K), 0.0076966 secs] 
[GC (Allocation Failure) 41380K->41476K(1067520K), 0.0083110 secs] 
[GC (Allocation Failure) 52740K->52620K(1068544K), 0.0101210 secs] 
[GC (Allocation Failure) 63884K->64060K(1079808K), 0.0126912 secs] 
[GC (Allocation Failure) 81468K->81436K(1080832K), 0.0131405 secs] 
[GC (Allocation Failure) 98844K->99052K(1093120K), 0.0189248 secs] 
[GC (Allocation Failure) 121580K->121684K(1095168K), 0.0154816 secs] 
[GC (Allocation Failure) 144212K->144308K(1105408K), 0.0233011 secs] 
[GC (Allocation Failure) 173492K->173556K(1107968K), 0.0199322 secs] 
[GC (Allocation Failure) 202740K->202860K(1120768K), 0.0276851 secs] 
[GC (Allocation Failure) 240236K->240396K(1124352K), 0.0240986 secs] 
[GC (Allocation Failure) 277772K->277908K(1140736K), 0.0352077 secs] 
[GC (Allocation Failure) 326036K->326180K(1145344K), 0.0325568 secs] 
[GC (Allocation Failure) 374308K->374484K(1161728K), 0.0419046 secs] 
[GC (Allocation Failure) 431316K->431548K(1167872K), 0.0383821 secs] 
[GC (Allocation Failure) 488380K->488580K(1187328K), 0.0488973 secs] 
[GC (Allocation Failure) 556676K->556860K(1193472K), 0.0427763 secs] 
[GC (Allocation Failure) 624956K->625220K(1216000K), 0.0556787 secs] 
[GC (Allocation Failure) 706628K->706852K(1223168K), 0.0524774 secs] 
[GC (Allocation Failure) 788260K->788492K(1249792K), 0.0675712 secs] 
Done 

特別是，有沒有什麼大的垃圾收集，程序中的CPU時間的一小部分完成。

當然，如果您在程序過程中反覆創建這樣的大型數組，那麼會發生垃圾回收，並且臨時使用其他內核可能仍是一個問題。在這種情況下，我建議通過其配置選項來調整垃圾收集器，以某種方式誘使JVM相信可用內核較少，或者在操作系統級別限制活動JVM線程的數量。

Answer 2

這裏你沒有明確使用任何多線程代碼，所以如果你看到多CPU使用率，其他線程可能是JVM內部，你無法控制它們。

Answer 3

您遇到的問題是Java不會執行原生多維數組。相反，你有一個由數組數組組成的數組數組。如果你使a = 1000和e = 10，它會快得多。