"Эффективность и оптимизация кода в Java 8" Сергей Моренец
Download as PPTX, PDF•3 likes•1,934 views
The document is a presentation by Sergey Morenets focusing on effectiveness and code optimization in Java applications, detailing topics such as JVM optimization, code measurement techniques, and benchmarking frameworks like JMH. It emphasizes the importance of avoiding premature optimization and highlights various coding techniques and tools for enhancing performance and efficiency. The presentation also includes code examples, performance measurements, and a discussion on different data structures and synchronization methods.
"Эффективность и оптимизация кода в Java 8" Сергей Моренец
- 1. Заголовок презентации Имя и фамилия автора доклада Компания Контактные данные 1 Контактные данные 2 Контактные данные 3 Контактные данные 4 Effectiveness and code optimization in Java applications Sergey Morenets [email protected]
- 2. About author • Works in IT since 2000 • 12 year of Java SE/EE experience • Regular speaker at Java conferences • Author of “Development of Java applications” and “Main errors in Java programming ”books • Founder of http://it-simulator.com
- 3. Preface
- 4. Agenda
- 5. Agenda • What is effectiveness? • Code optimization • JVM optimization • Code samples • Measurements
- 6. Ideal code Concise Readable Self-describing Reusable Testable Modern Flexible Scalable Effective
- 7. Effectiveness • Hard to determine on code/design review stages or in unit-tests • Is relevant for the specific project configuration/environment • Premature optimization is evil • Hardware-specific • The only aspect of the code that affects users
- 16. Effectiveness • Can be measured • Can be static or dynamic • Can be tuned
- 17. Tuning • JVM options • Metaspace/heap/stack size • Garbage collector options • http://blog.sokolenko.me/2014/11/javavm- options-production.html • http://www.javaspecialists.eu/
- 18. Code optimization Java compiler JIT compiler JVM
- 20. Code optimization public int execute(); Code: 0: iconst_2 1: istore_1 2: iinc 1, 1 5: iconst_1 6: ireturn
- 22. Code optimization public int execute(); Code: 0: iconst_2 1: istore_1 2: iconst_1 3: ireturn
- 24. Code optimization public void execute(); Code: 0: return
- 26. Code optimization public static boolean get(); Code: 0: iconst_1 1: ireturn
- 28. Code optimization public void execute(); Code: 0: return
- 30. Code optimization public int execute(); Code: 0: iconst_2 1: istore_1 2: iconst_4 3: istore_2 4: iload_1 5: iload_2 6: iadd 7: ireturn
- 32. Code optimization public int execute(); Code: 0: bipush 6 2: ireturn
- 34. Code optimization public int execute(); Code: 0: bipush 12 2: ireturn
- 35. Javacompiler Dead code elimination Constant folding Fixed expression calculation
- 36. Measurements • JMH is micro benchmarking framework • Developed by Oracle engineers in 2013 • Requires build tool(Maven, Gradle) • Can measure throughput or average time • Includes warm-up period
- 37. Warm-up
- 38. Environment • JMH 1.11.3 • Maven 3.3.9 • JDK 1.8.0.73 • Intel Core i7 4790, 4 cores, 16 GB
- 39. Measurements Type Time(ns) Multiply 4 2,025 Shift 2,024
- 40. Measurements Type Time(ns) Multiply 17 2,04 Shift 2,04
- 42. Method vs Field
- 44. Inversion
- 45. Inversion
- 46. Measurements Type Time(ns) ! 1,99 ? 1,99 ^ 2,01
- 47. Arrays
- 48. Measurements Type Time(ns) For (10 elements) 4,9 For-each (10 elements) 5,1 For (1000 elements) 260 For-each (1000 elements) 259,9 For (50000 elements) 12957 For-each (50000 elements) 12958
- 49. Arrays
- 50. Measurements Type Time(ns) For (10 elements) 5,04 For optimized(10 elements) 5,07 For (1000 elements) 258,9 For-each (1000 elements) 258,7
- 51. Arrays
- 52. Measurements Type Time(ns) Sequential (10 elements) 5 Parallel (10 elements) 6230 Sequential (1000 elements) 263 Parallel (1000 elements) 8688 Sequential (50000 elements) 13115 Parallel (50000 elements) 34695
- 53. Measurements Type Time(ns) Sequential (10 elements) 5 Parallel (10 elements) 6230 Sequential (1000 elements) 263 Parallel (1000 elements) 8688 Sequential (50000 elements) 13115 Parallel (50000 elements) 34695 Sequential (5 000 000 elements) 1 765 206 Parallel (5 000 000 elements) 2 668 564 Sequential (500 000 000) 183 ms Parallel (500 000 000) 174 ms
- 54. Autoboxing
- 55. Measurements Type Time(ns) Array(100 elements) 58 List(100 elements) 390 Array(10 000 elements) 4776 List(10 000 elements) 48449
- 56. LIFO
- 58. Stack
- 59. Java NIO
- 60. Measurements 100 elements Array Heap buffer Direct buffer Create 13,87 18,9 502,5 Get 2,24 3,18 2,65 Update 2,26 3,21 2,93 Update all 29,5 33,2 36,6
- 61. Measurements 10000 elements Array Heap buffer Direct buffer Create 544 548 1543 Get 2,25 3,26 2,77 Update 2,38 3,19 2,94 Update all 2701 2720 5022
- 62. Collections
- 63. Collections Type Time(ns) Fill HashMap(1000 elements) 16000 Fill TreeMap(1000 elements) 40115 Fill HashMap(100 000 elements) 2 027 116 Fill TreeMap(100 000 elements) 11 195 422 Iteration HashMap(1000 elements) 3086 Iteration TreeMap(1000 elements) 5038
- 64. Reflection
- 65. Measurements Type Time(ns) New object 3,0 Reflection 5,4
- 66. Reflection
- 67. Measurements Type Time(ns) Method call 0,3 Reflection 232
- 68. Measurements Type Time(ns) Method call 0,3 Reflection (cached) 3,1
- 70. Lists
- 71. Measurements Type Time(ns) ArrayList (1000 elements) 4766 ArrayList (100 000 elements) 381707 LinkedList (1000 elements) 5504 LinkedList (100 000 elements) 504231
- 72. Lists
- 73. Measurements Type Time(ns) ArrayList (1000 elements) 26767 ArrayList (100 000 elements) 276(ms) LinkedList (1000 elements) 300971 LinkedList (100 000 elements) 3424(ms)
- 74. Lists
- 75. Measurements Type Time(ns) ArrayList (1000 elements) 774 ArrayList (100 000 elements) 144814 LinkedList (1000 elements) 2161 LinkedList (100 000 elements) 292364
- 78. Measurements Type (elements) Time(ns) ArrayList (1000) 4732 ArrayList (100 000) 387692 LinkedList (1000) 5775 LinkedList (100 000) 511646 ObjectArrayList(1000) 3168 ObjectArrayList(100 000) 322811
- 80. Measurements Type(1 thread) Time(ns) Not synchronized 2,0 Synchronized method 18,3 Synchronized block 18,3
- 84. Measurements Type (1 thread) Time(ns) ReentrantLock 18,7 ReadWriteLock 19,0 StampedLock 18,6
- 85. Synchronization
- 86. Measurements Type (16 threads) Time(ns) ReentrantLock 54,2 ReadWriteLock 56,4 StampedLock 52,5 AtomicLong 8,8
- 87. Java 7
- 88. Java 8
- 89. Java 8
- 90. Measurements Type Time(ns) Anonymous class 2,03 Lambda expression 2,37 Method reference 2,38
- 91. Java 7
- 92. Java 8
- 93. Measurements Type(# elements) Time(ns) For-loop(10) 5,0 Stream(10) 35,2 For-loop(1000) 264 Stream(1000) 1970 For-loop(50 000) 13244 Stream(50 000) 95689
- 94. Parallel streams
- 95. Measurements Type(# elements) Time(ns) Sum sequential(100) 6 Sum parallel(100) 51 Sleep sequential(100) 101 106 306 Sleep parallel(100) 13 575 733 Sum sequential(10000) 19517 Sum parallel(10000) 15657 Sleep sequential(10000) 10 068 771 891 Sleep parallel(10000) 1 250 911 296
- 96. Default methods
- 98. JIT optimization • Inline methods • Eliminate locks • Replace interface with direct method calls • Join synchronized blocks • Eliminate dead code • Drop memory write for non-volatile variables
- 99. Conclusion • Compiler and JIT optimization • Speed and memory optimization • Prefer ArrayList/HashMap • Synchronization and reflection cost • Use measurement tools
- 100. Theory