The Aroma Project
Link to the Home Page of The National Institute of Standards and Technology Link to the Home Page of The Software Diagnostics and Conformance Testing Division Link to the Home Page of The Standards and Conformance Testing Group
 Aroma Site Index
Aroma Overview
Home
NIST Aroma Projector
NIST Aroma Projector
Aroma Shell
Projector Software on the Adapter
Feather
Projector Software on the Laptop
The Aroma Adapters
NIST Aroma Adapters
Aroma Adapter Native Performance
Performance Measurements
Measurement Insights
JavaTM Technology Performance on the Aroma Adapter
Measurements
Responsiveness Issues
Responsiveness Results
Aroma's Future
mAroma
 
 Responsiveness Results
Responsiveness Results (1)
Comparison graphic shows the Five Worst Times for Sleeping 1 second for Native and Green Threads running on a Pentium III with 20 users
Click on image for larger graphic.
  • Since the VolanoMark is based on the VolanoChat server, imagine being the poor sysadmin trying to log on when the server is really loaded down with native threads!
Responsiveness Results (2)
Comparison graphic shows the Five Worst Times for Sleeping 1 second for Native and Green Threads running on a Pentium III with 10 users
Click on image for larger graphic.
  • With less load, the periods of sluggishness due to the native threads are less frequent.
Responsiveness Results (3)
Comparison graphic shows the Five Worst Times for Sleeping 1 second for Native and Green Threads running on a Cyrix GXM 233 with 10 users
Click on image for larger graphic.
  • The lower performance processor is more sensitive to a large number of native threads. This is not what you would expect from just looking at the previous benchmark results.

 

Whatís Going On...
  • Native threads are scheduled by the OS.

    • For a process running at a given priority, more threads means a greater share of CPU time.
  • Green threads are scheduled by the application.

    • The overall process is scheduled by the OS.
    • Each process timeslice can be shared among several threads.
Measure Things Often
  • Native hardware performance measurements are not enough.
  • Need to routinely measure application performance during development.

    • Scenario to avoid: The performance of your multithreaded mega-service meets the requirements. But then you add a few remaining small services and while the mega- service runs about the same, your customers start to complain about response time.
Understand your VM
  • The VMís thread implementation can affect performance and scalability of your application.
  • Green threads use less memory and can make the system more responsive.

  • Sometimes you canít choose your thread implementation.

Java and all Java-based trademarks and logos are trademarks or registered trademarks of Sun Microsystems, Inc. in the United States and other countries.

Any mention of commercial products within NIST web pages is for information only; it does not imply recommendation or endorsement by NIST.

Contact Aroma Project Leader Alden Dima
Last Updated 09 August 2001 by Tony Cincotta