EFFECTS OF LOCALITY, CONTENT AND JAVA RUNTIME ON VIDEO PERFORMANCE

1
EFFECTS OF LOCALITY, CONTENT AND JAVA
RUNTIMEON VIDEO PERFORMANCE

• Vikram Chhabra, Akshay Kothare,
• Mark Claypool
• Computer Science Department
• Worcester Polytechnic Institute
• Worcester, Massachusetts, USA

2
Introduction

• Increased bandwidth has enabled video streaming
  to the desktop.
• Javas Write once, run anywhere - preferred
  choice for multimedia applications.

3
Java Runtime
Java Source
C Source
Bytecode
JIT Compiler
JVM
Native OS
4
Java Performance

• Java lags the speed of C / C.
• Need to explore factors influencing the
  performance of Java.

5
Performance of Multimedia

• Frame Rate Numbers of frames displayed per
  second.
• Jitter Variance in time taken for a frame to
  travel from server to client.

6
Previous Work

• Most performance evaluations have been for
  traditional media.
• Our previous work suggests
• Local access of media
• Processing power of the native hardware
• mostly influence the performance of Java.

7
Our Approach

• Setup a controlled environment
• Serve movie to Java client
• Observe frame arrival time
• Calculate frame rate jitter
• Derive conclusions

8
Experimental Setup
C Server
Java Client
TCP/IP No frame loss
Hard Disk
9
Parameters to be tested

• Network distance
• Movie content
• JIT compilation

10
Parameter One Network Distance

• Four Setups
• Local Playback
• Dedicated Client/Server
• Normal Setup
• Different Switch
• Fixed Parameters
• Operating System Linux 6.3 kernel version
  2.2.13.
• Server Pentium MMX 233 MHz PC with 64MB of RAM.
• Client Pentium II 300 MHz PC with 128MB of RAM.
• Network Ethernet broadband coaxial cable at
  10Mbps.
• Movie A space scene, encoded as 120 I frames

11
Parameter Two Movie Content

• Ten perceptually different movies
• fixed resolution of 200 x 200 pixels
• 120 frames each, 4 seconds each
• I and IBBPBBPBBPBB group of pictures format.
• I Frame Fully Encoded
• B or P Frames Relatively Encoded
• Fixed Parameters
• Client WinNT 4.0 service pack 6
• (Pentium MMX 233 MHz, 64MB RAM)
• Server SuSE Linux 6.4 Kernel version 2.2.14
• (Intel PIII 500Mhz PC, 128 MB RAM)
• Network 100 Mbps Ethernet

12
The Movies
13
Parameter Three JIT Compilation

• JIT on vs. off
• Java vs. C
• Fixed Parameters
• Operating System Linux 6.3 kernel version
  2.2.13.
• Client Pentium II 300 MHz PC with 128MB of RAM.
• Movie A space scene, encoded as 120 I
  frames

14
RESULTS
15
Network Distance Frame Rate
Local media access faster than different
network setups
16
Network Distance Jitter
Local media access shows lesser jitter than
network access
17
Movie Content Frame Rate
Movie content affects the frame rate
18
Movie Content Jitter
Movie content affects jitter
19
JIT Compilation Frame Rate
JIT enhances performance, Java still lags power
of C
20
JIT Compilation Jitter
JIT reduces jitter
21
Conclusions

• Local media access faster than network
• 5 higher frame rate, 25 lesser jitter
• Not much difference over different networks
• Movie content has significant implications
• Blue movie 3 times faster and shows 40 lesser
  jitter than Mixbag

22
Conclusions (contd.)

• JIT gives performance boost
• 100 increase in frame rate
• 25 decrease in jitter
• Java still lags C
• Java (JIT on) frame rate 10 times less than C

23
Future Work

• Trying different protocols
• UDP, RMI, CORBA instead of TCP.
• Finding relationship between decrease in frame
  rate with perceptually different movie content.
• Trying different JVMs and/or JITs.