Bug ID: JDK-4757212 JVM 1.4.2 regression on Sierra EE startup compared to JVM 1.3.1 in mixed mode

Type: Bug
Component: hotspot
Sub-Component: compiler
Affected Version: 1.4.1,1.4.2

Priority: P3
Status: Closed
Resolution: Won't Fix
OS: solaris_8
CPU: sparc

Submitted: 2002-10-03
Updated: 2003-10-02
Resolved: 2003-10-02

JVM 1.4.2 (client) has a regression on Sierra EE RC4 startup compared to JVM 1.3.1 with default HotSpot options: 34.56 sec vs 28.47 sec of total user CPU time (21% regression) and compared to JVM 1.4.1: 34.56 sec vs. 32.53 sec (6% regression). Regression in C++ function Compilation::compile_java_method() was observed.
To reproduce:
-Install JDK 1.3.1
-Install Sun ONE Studio Sierra EE using Installer with JDK 1.3.1 
-Run command: /bin/time <sierra_with_jdk1.3.1_install_path>/bin/runide.sh -J-Dnetbeans.close=true
-Install JDK 1.4.2 with the latest libjvm.so
-Install Sun ONE Studio Sierra EE using Installer with JDK 1.4.2 
-Run command: /bin/time <sierra_with_jdk1.4.2_install_path>/bin/runide.sh -J-Dnetbeans.close=true
Note: second and following startups were always measured.

###@###.### 2002-10-02

EVALUATION Some work has been done recently to improve startup time. ###@###.### has fixed 4641403 (Reduce use of TLS) which seems to bring a good startup time benefit as well as improving run-time performance. ###@###.### has been working on compile speed improvements for the HotSpot client compiler (4741220). This work should make it in to 1.4.2. Outside the VM, I have been working with a few people including ###@###.###, ###@###.###, and ###@###.### on optimizations to the JDK core libraries: see 4747351, 4747353, 4747354, 4747357, 4747358, 4747361, 4747362, and 4747365. These have been put back and should be present in JDK 1.4.2 build 04. Scott has done some measurement with Netbeans and sees a pretty good improvement to startup time with these optimizations. More work in this area is ongoing. We'll keep this bug open to see whether these improvements address the Netbeans startup time concerns. ###@###.### 2002-10-04 The NetBeans group has provided some startup time numbers for NetBeans version 3.4: jdk1.3.1_05 21.8 21.3 jdk1.4.1fcs 31.1 31.1 jdk1.4.2b6 27.8 28.2 The improvements in 1.4.2 b05 had a positive impact but there is still a startup time regression relative to 1.3.1. It is unlikely that we will make significant additional startup time gains in 1.4.2 as all of the current proposals for doing so have broad impact and are only being considered for 1.5. ###@###.### 2002-11-13 It's unlikely that it will be possible to do any more work in 1.4.2 on startup time so this bug is being committed to 1.5. ###@###.### 2002-11-21 The issues covered in the bug seem to be adequately addressed by other bugs already in the system, so I'm closing this bug out. ###@###.### 2003-10-02

21-11-2002

Duplicate :	JDK-4757208 - JVM 1.4.1 regression on Sierra EE startup compared to JVM 1.3.1 in mixed mode
Relates :	JDK-4747354 - Use ASCIICaseInsensitiveComparator for locale tables
Relates :	JDK-4747361 - Optimize Character for US-ASCII
Relates :	JDK-4747353 - Cache filename canonicalization results
Relates :	JDK-4747362 - Optimize CharacterEncoding for US-ASCII
Relates :	JDK-4747357 - Logging should be initialized lazily
Relates :	JDK-4758432 - JVM 1.4.2 regression on Sierra EE startup compared to JVM 1.3.1 with -Xcomp
Relates :	JDK-4747351 - Compile .properties files into .class files
Relates :	JDK-4747358 - WindowsPreferences should initialize logging lazily
Relates :	JDK-4757211 - JVM 1.4.1 regression on Sierra EE startup compared to JVM 1.3.1 with -Xcomp
Relates :	JDK-4741220 - improve compilation speed
Relates :	JDK-4747365 - Avoid Manifest parsing