Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid...

20
www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved. www.mimos.my Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance Improvement

Transcript of Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid...

Page 1: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.www.mimos.my

Nazarudin WijeeMohd Sidek SallehGrid Computing LabMIMOS Berhad

P-GRADE Performance Improvement

Page 2: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

• Previously, system response time was slow whenever user access the P-GRADE concurrently.

• Users expecting the page response time should be faster then current response time even if user access it concurrently.

• Figure below illustrate the current average page response time for workflow monitoring and workflow submission .

Background

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

Page 3: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

Goal

Continued..

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

1 2 3 4 5 Ave0.00

20,000.00

40,000.00

60,000.00

80,000.00

100,000.00

120,000.00

Page Response Time - Workflow Monitoring Before Improvement

Test Execution Number

Ave Response Time (ms)

1 2 3 4 5 Ave0.00

1,000.00

2,000.00

3,000.00

4,000.00

5,000.00

Page Response Time - Workflow Submission Before Improvement

Test Execution Number

Ave Response Time (ms)

Functionality Average Response Time (s)

Maximum Response Time (s)

Workflow Monitoring < 5 < 10

Workflow Submission < 5 < 10

Page 4: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

1. Reduce the number of Cascading Style Sheet (CSS) used by the portal to improve the page loading time

2. Reduce the page size.3. Minimize the Log4J logging4. Configure the Hibernate parameter5. Configure the Linux File Descriptor parameter6. Redefine the doSubmit.sh job execution script7. Change database (mysql instead of hsql)8. Change the hardware specification

P-GRADE Portal Improvements

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

Page 5: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

Before Tuning:@import url(general.css);@import url(page.css);@import url(tab.css);@import url(window.css);@import url(portlet.css);@import url(layout.css);@import url(ui.css);

1. Reduce the number of CSS loading of web page (I) – default.css

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

After Tuning:a { color: #000000; text-decoration: none;}a:link { color: #0D70C6; text-decoration: bold;}a:visited { color: #0D70C6; text-decoration: none;}a:active { color: blue;}……

Default.css is the file that controls the look-and-feel of the Portal page

Combined 7 css files into single file

Page 6: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

Login Form page:

2. Reduce the page size – Login Form

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

250 KB 135KB

Page 7: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

Log4j.properties controls how the application display logging information.

3. Minimizing Log4J logging

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

Before Tuning:

Log4j.debug=TRUElog4j.logger.org.hsqldb=DEBUGlog4j.logger.org.gridlab.gridsphere.core.persist

ence.hibernate=DEBUGlog4j.logger.org.gridlab.gridsphere=DEBUG

After Tuning:

log4j.debug=FALSE#log4j.logger.org.hsqldb=DEBUG#log4j.logger.org.gridlab.gridsphere.core.persisten

ce.hibernate=DEBUG#log4j.logger.org.gridlab.gridsphere=DEBUG

4. Hibernate framework Tuning Hibernate.properties defines the behavior of Hibernate framework

After Tuning:

hibernate.c3p0.max_size 200hibernate.c3p0.min_size 100hibernate.c3p0.acquire_increment 5

Before Tuning:

hibernate.c3p0.max_size 50hibernate.c3p0.min_size 30hibernate.c3p0.acquire_increment 2

Page 8: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

Limits.conf defines the resource usage limitation for the Linux user.

5. Configure Linux File Descriptor Parameter

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

Before Tuning:

* soft nofile 1024

* hard nofile 1024

After Tuning:

* soft nofile 9216

* hard nofile 9216

6. Redefine doSubmit.sh job execution scriptBefore Tuning: After Tuning:

SZGWorkflow.java

doSubmit.sh

SZGWorkflow.java

doSubmitWrap.sh#!/bin/bash.....doSubmit.sh $1 $2 >> $LOG &echo 1 > $WORKFLOW_DIR/lock

if [ -z "$CATALINA_BASE" ] ; then CATALINA_BASE="$CATALINA_HOME"fi

if [ -z "$CATALINA_TMPDIR" ] ; then # Define the java.io.tmpdir to use for Catalina CATALINA_TMPDIR="$CATALINA_BASE"/tempfi

# Bugzilla 37848: When no TTY is available, don't output to consolehave_tty=0if [ "`tty`" != "not a tty" ]; then have_tty=1

Page 9: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

7. Changing the database

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

Before Tuning:

HSQL

After Tuning:

MySQL

8. Changing the Server Hardware Specification

Before Tuning:

CPU 2.8 GHzAMD Opteron Duo Core1GB RAM 38G Harddisk

After Tuning:

CPU 2.8 GHz2 x AMD Opteron Duo Core16GB RAM127GB Harddisk

Page 10: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

100 concurrent virtual users with 0 delay between users

P-Grade Performance – Workflow Monitoring After Improvement

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

Test Execution Number

Page 11: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

100 concurrent virtual users with 0 delay between users

P-Grade Performance – Workflow Submission After Improvement

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

Test Execution Number

Test Execution Number

Test Execution Number

Page 12: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

Job Management - Control Chart

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

o RPT is used to automate and simulate 100 concurrent users with 0 Delay (ms) doing Job Management. The chart shows Job Management mean for 10 iterations.

o The 10 observations for both before and after show within control limit. Process is stable.

Page 13: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

Job Management - Comparison

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

o The mean was brought down by 1491.74 ms after the performance improvement.

Page 14: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

Job Management - Hypothesis Testing

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

Two-sample T: Ho: Job Management mean BEFORE improvement <= Job Management mean AFTER improvement Ha: Job Management mean BEFORE improvement > Job Management mean AFTER improvement

P value = 0.000 Reject Null Hypothesis

Conclusion • This is to conclude that both Job Management mean - with and without performance improvement do not perform equally.

• Performance improvement in P-Grade Portal proved response time reduction significantly on Job Management.

Page 15: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

Job Submission Time - Control Chart

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

o RPT is used to automate and simulate 100 concurrent users with 0 Delay (ms) doing job submission. The chart shows job submission mean for 10 iterations.

o The 10 observations for both before and after show within control limit.

Page 16: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

Job Submission Time - Comparison

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

o The mean was brought down by 52732.28 ms after the performance improvement.

Page 17: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

Job Submission Time - Hypothesis Testing

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

Two-sample T: Ho: Job Submission mean BEFORE improvement <= Job Submission mean AFTER improvement Ha: Job Submission mean BEFORE improvement > Job Submission mean AFTER improvement

P value = 0.000 Reject Null Hypothesis

Conclusion • This is to conclude that both Job Submission mean - with and without performance improvement do not perform equally.

• Performance improvement in P-Grade Portal proved response time reduction significantly on job submission.

Page 18: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

Q10: What aspects of my project have been quantified and stabilized with which I can demonstrate prediction of process capability?

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

• Performance improvement in P-Grade Portal proved response time reduction significantly on job submission. Job submission process AFTER the improvement is proved to be MUCH CAPABLE.

Page 19: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

Continued..

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.

• Performance improvement in P-Grade Portal proved response time reduction significantly on job monitoring. Job submission process AFTER the improvement is proved to be MUCH CAPABLE.

Page 20: Www.mimos.my© 2010 MIMOS Berhad. All Rights Reserved. Nazarudin Wijee Mohd Sidek Salleh Grid Computing Lab MIMOS Berhad P-GRADE Performance.

THANK YOU

www.mimos.my © 2010 MIMOS Berhad. All Rights Reserved.