It’s good to see works in progress for Aqua/Cocoa - AWT ports but what about JVM performances ?
Test vms
I selected 4 VMs to be tested
Apple Java 1.6.0_22 - java version "1.6.0_22" Java(TM) SE Runtime Environment (build 1.6.0_22-b04-314-10M3406a) Java HotSpot(TM) 64-Bit Server VM (build 17.1-b03-314, mixed mode)
Apple Java 1.6.0_24 - java version "1.6.0_24" Java(TM) SE Runtime Environment (build 1.6.0_24-b07-348-10M3406a) Java HotSpot(TM) 64-Bit Server VM (build 19.1-b02-348, mixed mode)
OpenJDK 7 bsd-port - openjdk version "1.7.0-internal" OpenJDK Runtime Environment (build 1.7.0-internal-henri_2011_04_11_08_24-b00) OpenJDK 64-Bit Server VM (build 21.0-b07, mixed mode)
OpenJDK 7 macosx-port - openjdk version "1.7.0-internal" OpenJDK Runtime Environment (build 1.7.0-internal-b00) OpenJDK 64-Bit Server VM (build 21.0-b07, mixed mode)
Test system
My test system is an Apple iMac (iMac11,1 ) with Intel i7 2.80Ghz and 8Gb DDR3 1067Mhz, running under SnowLeopard 10.6.7 64bits. I wanted to test 64bits VMs on a 64bits machine and this time use a stronger processor with more threads (ie: 4 cores with hyperthreading).
DaCapo Benchmarks
I keep the DaCapo 9.12-bach.
Bench tests launched with -n X, ie (java -jar dacapo-9.12-bach.jar -n 10 pmd)
| Bench | Apple JDK6 b22 | Apple JDK6 b24 | OpenJDK 1.7 bsd-port | OpenJDK 1.7 macosx-port |
|---|---|---|---|---|
| avrora (10 iterations) | 3464ms | 3406ms | 3281ms | 3410ms |
| eclipse (2 iterations) | 25635ms | 23264ms | 22156ms | 23503ms |
| fop (10 iterations) | 379ms | 351ms | 301ms | 305ms |
| h2 (2 iterations) | 5662ms | 5308ms | 4557ms | 4694ms |
| jython (2 iterations) | 4287ms | 4188ms | Failure (Trace/BPT trap) | 4004ms |
| luindex (10 iterations) | 2402ms | 763ms | 623ms | 670ms |
| lusearch (10 iterations) | 1500ms | 2173ms | 1190ms | 4019ms |
| pmd (10 iterations) | 2054ms | 1860ms | 1582ms | 1891ms |
| sunflow (10 iterations) | 2763ms | 2658ms | 2342ms | 2292ms |
| tomcat (5 iterations) | 1943ms | 1884ms | 1653ms | 1778ms |
| tradebeans (5 iterations) | 6702ms | 6199ms | 4968ms | 5080ms |
| tradesoap (5 iterations) | 20058ms | 18501ms | 21850ms | 20114ms |
| xalan (10 iterations) | 1080ms | 926ms | 788ms | 805ms |
Conclusion
Latest Apple JVM, 1.6.0-24 perform better than the old 1.6.0-22 in all of the tests and is near OpenJDK 7 results.
OpenJDK 7 from the bsd-port perform a little better than the macosx-port. The main difference in build is bsd-port is using stock gcc whereas macos-port use llvm-gcc.
bsd-port using stock-gcc during OpenJDK build
Compiling /Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-bsdport-x86_64/workspace/hotspot/src/share/vm/adlc/arena.cpp rm -f ../generated/adfiles/arena.o /Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-bsdport-x86_64/workspace/ALT_COMPILER_PATH/g++ -D_ALLBSD_SOURCE -D_GNU_SOURCE -DAMD64 -I/Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-bsdport-x86_64/workspace/hotspot/src/share/vm/prims -I/Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-bsdport-x86_64/workspace/hotspot/src/share/vm -I/Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-bsdport-x86_64/workspace/hotspot/src/cpu/x86/vm -I/Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-bsdport-x86_64/workspace/hotspot/src/os_cpu/bsd_x86/vm -I/Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-bsdport-x86_64/workspace/hotspot/src/os/bsd/vm -I/Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-bsdport-x86_64/workspace/hotspot/src/os/posix/vm -I/Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-bsdport-x86_64/workspace/hotspot/src/share/vm/adlc -I../generated -DASSERT -DTARGET_OS_FAMILY_bsd -DTARGET_ARCH_x86 -DTARGET_ARCH_MODEL_x86_64 -DTARGET_OS_ARCH_bsd_x86 -DTARGET_OS_ARCH_MODEL_bsd_x86_64 -DTARGET_COMPILER_gcc -DCOMPILER2 -DCOMPILER1 -fno-rtti -fno-exceptions -pthread -fcheck-new -m64 -pipe -Werror -g -c -o ../generated/adfiles/arena.o /Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-bsdport-x86_64/workspace/hotspot/src/share/vm/adlc/arena.cpp
macosx-port using llvm-gcc during OpenJDK build
Compiling /Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-macosx-universal/workspace/hotspot/src/share/vm/adlc/arena.cpp rm -f ../generated/adfiles/arena.o llvm-g++ -D_ALLBSD_SOURCE -D_GNU_SOURCE -DIA32 -I/Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-macosx-universal/workspace/hotspot/src/share/vm/prims -I/Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-macosx-universal/workspace/hotspot/src/share/vm -I/Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-macosx-universal/workspace/hotspot/src/cpu/x86/vm -I/Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-macosx-universal/workspace/hotspot/src/os_cpu/bsd_x86/vm -I/Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-macosx-universal/workspace/hotspot/src/os/bsd/vm -I/Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-macosx-universal/workspace/hotspot/src/os/posix/vm -I/Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-macosx-universal/workspace/hotspot/src/share/vm/adlc -I../generated -DASSERT -DTARGET_OS_FAMILY_bsd -DTARGET_ARCH_x86 -DTARGET_ARCH_MODEL_x86_32 -DTARGET_OS_ARCH_bsd_x86 -DTARGET_OS_ARCH_MODEL_bsd_x86_32 -DTARGET_COMPILER_gcc -DCOMPILER2 -DCOMPILER1 -fno-rtti -fno-exceptions -pthread -fcheck-new -m32 -march=i586 -mstackrealign -pipe -Werror -g -c -o ../generated/adfiles/arena.o /Users/henri/Documents/jenkins/data/jobs/openjdk-1.7-macosx-universal/workspace/hotspot/src/share/vm/adlc/arena.cpp
Performances gain in OpenJDK7 VM vs latest Apple 6 VM is smaller than previously (see previous articles on Apple JDK vs OpenJDK 6), switching to OpenJDK 7 will not be only for pure speed but for functionalities.