HotSpotコトハジメ
•Download as PPTX, PDF•
25 likes•7,294 views
- The document discusses optimizing Java performance on hardware by leveraging features like AES-NI, transparent huge pages, and compiler intrinsics. - It provides examples showing performance improvements from using these features, such as faster encryption times when enabling AES-NI and fewer TLB misses with transparent huge pages. - It also discusses Java VM options, garbage collection, and the HotSpot compiler related to hardware optimization.
![• CodeCacheはGCでクリアされません！
•
•
•
ちょっと
だけ
Java HotSpot(TM) 64-Bit Server VM warning: CodeCache is full. Compiler has been disabled.
Java HotSpot(TM) 64-Bit Server VM warning: Try increasing the code cache size using -XX:ReservedCodeCacheSize=
CodeCache: size=2496Kb used=1955Kb max_used=1969Kb free=540Kb
bounds [0x00007f50c844b000, 0x00007f50c86bb000, 0x00007f50c86bb000]
total_blobs=578 nmethods=353 adapters=139
compilation: disabled (not enough contiguous free space left)](https://image.slidesharecdn.com/hotspot-150411082239-conversion-gate01/85/HotSpot-18-320.jpg)
![•
•
•
•
# cat /sys/kernel/mm/transparent_hugepage/enabled
always [madvise] never](https://image.slidesharecdn.com/hotspot-150411082239-conversion-gate01/85/HotSpot-26-320.jpg)
![import java.nio.*;
import java.nio.file.*;
import javax.crypto.*;
import javax.crypto.spec.*;
import java.security.*;
public class AESTest{
public static void main(String[] args) throws Exception{
Key key = new SecretKeySpec("passwordpassword".getBytes(), "AES");
Cipher cipher = Cipher.getInstance("AES/CBC/NoPadding");
cipher.init(Cipher.ENCRYPT_MODE, key);
byte[] plain = Files.readAllBytes(Paths.get(args[0]));
long start = System.currentTimeMillis();
cipher.doFinal(plain);
long end = System.currentTimeMillis();
long elapsedTime = end - start;
System.out.println("Elapsed time: " + elapsedTime + " ms");
}
}](https://image.slidesharecdn.com/hotspot-150411082239-conversion-gate01/85/HotSpot-27-320.jpg)
![import java.nio.*;
import java.nio.file.*;
import javax.crypto.*;
import javax.crypto.spec.*;
import java.security.*;
public class AESTest{
public static void main(String[] args) throws Exception{
Key key = new SecretKeySpec("passwordpassword".getBytes(), "AES");
Cipher cipher = Cipher.getInstance("AES/CBC/NoPadding");
cipher.init(Cipher.ENCRYPT_MODE, key);
byte[] plain = Files.readAllBytes(Paths.get(args[0]));
long start = System.currentTimeMillis();
cipher.doFinal(plain);
long end = System.currentTimeMillis();
long elapsedTime = end - start;
System.out.println("Elapsed time: " + elapsedTime + " ms");
}
}](https://image.slidesharecdn.com/hotspot-150411082239-conversion-gate01/85/HotSpot-45-320.jpg)
![https://wiki.openjdk.java.net/display/Sumatra/Standalone+Sumatra+Stream+API+Offload+Demo
final int length = 8;
int[] ina = new int[length];
int[] inb = new int[length];
int[] out = new int[length];
// Initialize the input arrays - this is offloadable
IntStream.range(0, length).parallel().forEach(p -> {
ina[p] = 1;
inb[p] = 2;
});
// Sum each pair of elements into out[] - this is offloadable
IntStream.range(0, length).parallel().forEach(p -> {
out[p] = ina[p] + inb[p];
});](https://image.slidesharecdn.com/hotspot-150411082239-conversion-gate01/85/HotSpot-53-320.jpg)
![•
•
•
•
Current thread (0x00007fb944008800): JavaThread "main" [_thread_in_native, id=10054, stack(0x00007fb9486b0000,0x00007fb9487b1000)]
siginfo: si_signo: 11 (SIGSEGV), si_code: 1 (SEGV_MAPERR), si_addr: 0x0000000000000000](https://image.slidesharecdn.com/hotspot-150411082239-conversion-gate01/85/HotSpot-65-320.jpg)
![•
•
Native frames: (J=compiled Java code, j=interpreted, Vv=VM code, C=native code)
C [libsegv.so+0x6a0] Java_NativeSEGV_doSEGV+0x18
j NativeSEGV.doSEGV()V+0
j NativeSEGV.main([Ljava/lang/String;)V+0
v ~StubRoutines::call_stub
V [libjvm.so+0x681a56] JavaCalls::call_helper(JavaValue*, methodHandle*, JavaCallArguments*, Thread*)+0x1056
V [libjvm.so+0x6c36c2] jni_invoke_static(JNIEnv_*, JavaValue*, _jobject*, JNICallType, _jmethodID*, JNI_ArgumentPusher*, Thread*)+0x362
V [libjvm.so+0x6e00ca] jni_CallStaticVoidMethod+0x17a
C [libjli.so+0x7bcc] JavaMain+0x80c
C [libpthread.so.0+0x752a]](https://image.slidesharecdn.com/hotspot-150411082239-conversion-gate01/85/HotSpot-66-320.jpg)
HotSpotコトハジメ
- 2. 今日はGCの話をあまりしません！ 他の方の資料等をご覧ください… OracleとJavaは、Oracle Corporation及びその子会社、関連会社の米国及びその他の国における登録商標です。 文中の社名、商品名等は各社の商標または登録商標である場合があります。 本資料の実装確認はOpenJDK8 8u-dev（HotSpot rev. 7164:bf68e15dc8fe）で行っています。
- 8. $ java -version openjdk version "1.8.0_31" OpenJDK Runtime Environment (build 1.8.0_31-b13) OpenJDK 64-Bit Server VM (build 25.31-b07, mixed mode) $ java -version java version "1.8.0" Java(TM) SE Runtime Environment (build 1.8.0-b132) Java HotSpot(TM) Client VM (build 25.0-b70, mixed mode)
- 11. • • • http://docs.oracle.com/javase/jp/8/technotes/guides/vm/s erver-class.html 注: Java SE 6の場合、server-classマシンとは、少なくとも2つのCPU と、最少でも2Gバイトの物理メモリーを備えているマシンを指し ます。
- 13. • • • •
- 14. • • • • • •
- 18. • CodeCacheはGCでクリアされません！ • • • ちょっと だけ Java HotSpot(TM) 64-Bit Server VM warning: CodeCache is full. Compiler has been disabled. Java HotSpot(TM) 64-Bit Server VM warning: Try increasing the code cache size using -XX:ReservedCodeCacheSize= CodeCache: size=2496Kb used=1955Kb max_used=1969Kb free=540Kb bounds [0x00007f50c844b000, 0x00007f50c86bb000, 0x00007f50c86bb000] total_blobs=578 nmethods=353 adapters=139 compilation: disabled (not enough contiguous free space left)
- 19. • • • •
- 20. • • • http://www.oracle.com/webfolder/technetwork/jp/javama gazine/Java-JA13-Architect-evans.pdf
- 22. • • • • $ ulimit -v 102400 $ /usr/local/jdk1.8.0/bin/java -version # # There is insufficient memory for the Java Runtime Environment to continue. # Native memory allocation (malloc) failed to allocate 80 bytes for AllocateHeap # An error report file with more information is saved as: # /home/xxx/hs_err_pid8664.log
- 23. • • • • • http://www.oracle.com/technetwork/jp/articles/java/compressedoops- 427542-ja.html • • •
- 24. • • • • • • •
- 25. • • • • • •
- 26. • • • • # cat /sys/kernel/mm/transparent_hugepage/enabled always [madvise] never
- 27. import java.nio.*; import java.nio.file.*; import javax.crypto.*; import javax.crypto.spec.*; import java.security.*; public class AESTest{ public static void main(String[] args) throws Exception{ Key key = new SecretKeySpec("passwordpassword".getBytes(), "AES"); Cipher cipher = Cipher.getInstance("AES/CBC/NoPadding"); cipher.init(Cipher.ENCRYPT_MODE, key); byte[] plain = Files.readAllBytes(Paths.get(args[0])); long start = System.currentTimeMillis(); cipher.doFinal(plain); long end = System.currentTimeMillis(); long elapsedTime = end - start; System.out.println("Elapsed time: " + elapsedTime + " ms"); } }
- 28. $ java -Xms2g -Xmx2g AESTest /tmp/test.dat Elapsed time: 1654 ms $ java -Xms2g -Xmx2g -XX:+UseTransparentHugePages AESTest /tmp/test.dat Elapsed time: 1613 ms 約40msの差！ ソフトウェア環境： • Fedora21 x86_64 • kernel-3.18.7-200.fc21.x86_64 • glibc-2.20-8.fc21.x86_64 • java-1.8.0-openjdk-1.8.0.31-5.b13.fc21.x86_64 ハードウェア環境： • CPU：Intel Core i5 2500T • MB：GIGABYTE GA-H67A-UD3H-B3 • メモリ：DDR3 1333MHz 2GB×2 テストデータ： • /tmp/test.dat • /dev/urandomから500MBを/tmp配下にコピー • /tmpはtmpfs
- 29. $ perf stat -e dTLB-load-misses,dTLB-store-misses java -Xms2g -Xmx2g AESTest /tmp/test.dat Elapsed time: 1654 ms Performance counter stats for 'java -Xms2g -Xmx2g AESTest /tmp/test.dat': 826,055 dTLB-load-misses 1,648,456 dTLB-store-misses 2.300445236 seconds time elapsed $ perf stat -e dTLB-load-misses,dTLB-store-misses java -Xms2g -Xmx2g -XX:+UseTransparentHugePages AESTest /tmp/test.dat Elapsed time: 1613 ms Performance counter stats for 'java -Xms2g -Xmx2g -XX:+UseTransparentHugePages AESTest /tmp/test.dat': 623,795 dTLB-load-misses 576,097 dTLB-store-misses 2.165985041 seconds time elapsed TLBミスヒットが激減！
- 30. • • • • • •
- 32. • • •
- 33. 論理プロセッサ数が8以下のとき 論理プロセッサ数が9以上のとき 論理プロセッサ数と同数 8 + 5 < 論理プロセッサ数 > −8 8
- 35. • • • • • • • •
- 37. import sun.misc.Contended; @Contended public class Test{ public int val1; public int val2; public int val3; public int val4; public int val5; }
- 38. $ java -cp jol-cli-0.3.1-full.jar:. org.openjdk.jol.Main internals Test ： Test object internals: OFFSET SIZE TYPE DESCRIPTION VALUE ： 12 4 int Test.val1 0 16 4 int Test.val2 0 20 4 int Test.val3 0 24 4 int Test.val4 0 28 4 int Test.val5 0 Instance size: 32 bytes (reported by Instrumentation API) Space losses: 0 bytes internal + 0 bytes external = 0 bytes total $ java -XX:-RestrictContended -cp jol-cli-0.3.1-full.jar:. org.openjdk.jol.Main internals Test ： Test object internals: OFFSET SIZE TYPE DESCRIPTION VALUE ： 12 128 (alignment/padding gap) N/A 140 4 int Test.val1 0 144 4 int Test.val2 0 148 4 int Test.val3 0 152 4 int Test.val4 0 156 4 int Test.val5 0 160 128 (loss due to the next object alignment) Instance size: 288 bytes (reported by Instrumentation API) Space losses: 128 bytes internal + 128 bytes external = 256 bytes total @Contended無効 @Contended有効 ソフトウェア環境： • Fedora21 x86_64 • kernel-3.18.7-200.fc21.x86_64 • glibc-2.20-8.fc21.x86_64 • java-1.8.0-openjdk-1.8.0.31-5.b13.fc21.x86_64
- 39. import sun.misc.Contended; public class Test2{ @Contended("group") public int val1; @Contended("group") public int val2; public int val3; public int val4; @Contended public int val5; }
- 40. $ java -cp jol-cli-0.3.1-full.jar:. org.openjdk.jol.Main internals Test2 ： Test2 object internals: OFFSET SIZE TYPE DESCRIPTION VALUE ： 12 4 int Test2.val1 0 16 4 int Test2.val2 0 20 4 int Test2.val3 0 24 4 int Test2.val4 0 28 4 int Test2.val5 0 Instance size: 32 bytes (reported by Instrumentation API) Space losses: 0 bytes internal + 0 bytes external = 0 bytes total $ java -XX:-RestrictContended -cp jol-cli-0.3.1-full.jar:. org.openjdk.jol.Main internals Test2 ： Test2 object internals: OFFSET SIZE TYPE DESCRIPTION VALUE ： 12 4 int Test2.val3 0 16 4 int Test2.val4 0 20 128 (alignment/padding gap) N/A 148 4 int Test2.val5 0 152 128 (alignment/padding gap) N/A 280 4 int Test2.val1 0 284 4 int Test2.val2 0 288 128 (loss due to the next object alignment) Instance size: 416 bytes (reported by Instrumentation API) Space losses: 256 bytes internal + 128 bytes external = 384 bytes total @Contended無効 @Contended有効 ソフトウェア環境： • Fedora21 x86_64 • kernel-3.18.7-200.fc21.x86_64 • glibc-2.20-8.fc21.x86_64 • java-1.8.0-openjdk-1.8.0.31-5.b13.fc21.x86_64
- 45. import java.nio.*; import java.nio.file.*; import javax.crypto.*; import javax.crypto.spec.*; import java.security.*; public class AESTest{ public static void main(String[] args) throws Exception{ Key key = new SecretKeySpec("passwordpassword".getBytes(), "AES"); Cipher cipher = Cipher.getInstance("AES/CBC/NoPadding"); cipher.init(Cipher.ENCRYPT_MODE, key); byte[] plain = Files.readAllBytes(Paths.get(args[0])); long start = System.currentTimeMillis(); cipher.doFinal(plain); long end = System.currentTimeMillis(); long elapsedTime = end - start; System.out.println("Elapsed time: " + elapsedTime + " ms"); } }
- 46. $ java -Xms2g -Xmx2g AESTest /tmp/test.dat Elapsed time: 1669 ms $ java -Xms2g -Xmx2g -XX:-UseAES AESTest /tmp/test.dat Elapsed time: 4651 ms ソフトウェア環境： • Fedora21 x86_64 • kernel-3.18.7-200.fc21.x86_64 • glibc-2.20-8.fc21.x86_64 • java-1.8.0-openjdk-1.8.0.31-5.b13.fc21.x86_64 約3秒差！ テストデータ： • /tmp/test.dat • /dev/urandomから500MBを/tmp配下にコピー • /tmpはtmpfs ハードウェア環境： • CPU：Intel Core i5 2500T • MB：GIGABYTE GA-H67A-UD3H-B3 • メモリ：DDR3 1333MHz 2GB×2
- 47. $ java -XX:+UnlockDiagnosticVMOptions -XX:+PrintIntrinsics AESTest /tmp/test.dat @ 64 com.sun.crypto.provider.AESCrypt::encryptBlock (953 bytes) (intrinsic) @ 80 java.lang.System::arraycopy (0 bytes) (intrinsic) Elapsed time: 1669 ms $ java -XX:+UnlockDiagnosticVMOptions -XX:+PrintIntrinsics -XX:-UseAES AESTest /tmp/test.dat @ 80 java.lang.System::arraycopy (0 bytes) (intrinsic) Elapsed time: 4651 ms
- 49. • • • • • • • •
- 53. https://wiki.openjdk.java.net/display/Sumatra/Standalone+Sumatra+Stream+API+Offload+Demo final int length = 8; int[] ina = new int[length]; int[] inb = new int[length]; int[] out = new int[length]; // Initialize the input arrays - this is offloadable IntStream.range(0, length).parallel().forEach(p -> { ina[p] = 1; inb[p] = 2; }); // Sum each pair of elements into out[] - this is offloadable IntStream.range(0, length).parallel().forEach(p -> { out[p] = ina[p] + inb[p]; });
- 54. We cannot speak for Project Sumatra or AMD. Project Graal decided to temporarily remove the compiler's GPU support, because the original contributors indicated to no longer maintain the code. We are still investigating outside the main branch a new version of compilation support for heterogenous architectures for Graal. If you want to compile and run a Graal version with GPU support in the meantime, use this revision: http://hg.openjdk.java.net/graal/graal/rev/91fee1fab96d <http://hg.openjdk.java.net/graal/graal/rev/91fee1fab96d>. http://mail.openjdk.java.net/pipermail/sumatra-dev/2015-February/000308.html
- 59. • • • http://www.oracle.com/webfolder/technetwork/jp/javama gazine/Java-JA12-architect-bajaj.pdf
- 63. • • •
- 65. • • • • Current thread (0x00007fb944008800): JavaThread "main" [_thread_in_native, id=10054, stack(0x00007fb9486b0000,0x00007fb9487b1000)] siginfo: si_signo: 11 (SIGSEGV), si_code: 1 (SEGV_MAPERR), si_addr: 0x0000000000000000
- 66. • • Native frames: (J=compiled Java code, j=interpreted, Vv=VM code, C=native code) C [libsegv.so+0x6a0] Java_NativeSEGV_doSEGV+0x18 j NativeSEGV.doSEGV()V+0 j NativeSEGV.main([Ljava/lang/String;)V+0 v ~StubRoutines::call_stub V [libjvm.so+0x681a56] JavaCalls::call_helper(JavaValue*, methodHandle*, JavaCallArguments*, Thread*)+0x1056 V [libjvm.so+0x6c36c2] jni_invoke_static(JNIEnv_*, JavaValue*, _jobject*, JNICallType, _jmethodID*, JNI_ArgumentPusher*, Thread*)+0x362 V [libjvm.so+0x6e00ca] jni_CallStaticVoidMethod+0x17a C [libjli.so+0x7bcc] JavaMain+0x80c C [libpthread.so.0+0x752a]
- 68. • • • • • • ： compile java/lang/invoke/MethodHandle <clinit> ()V -1 4 inline 1 0 -1 java/lang/invoke/MethodHandle <clinit> ()V
- 69. • Current CompileTask: C2: 13427 2 !b 4 java.lang.invoke.MethodHandle::<clinit> (75 bytes)
- 70. • • • • • •