Runtime optimization and monitoring of parallel visualization streams

1
Run-time optimization and monitoring of parallel
visualization streams

• Jianli Shen
• Samantika Subramaniam

2
Introduction

• Dproc
• -kernel-level, scalable monitoring tool kit
  for cluster applications
• SmartPointer
• -paradigm for utilizing multiple light-weight
  client endpoints in a real-time scientific
  visualization infrastructure.

3
Project goal

• Extending the current implementation of dproc
  (distributed /proc) to use it in Smartpointer to
  make for more efficient resource monitoring
• The service provided by the server should
  dynamically adapt itself
• Hence a fully distributed ,resource aware peer to
  peer system

4
Current Implementation of SmartPointer

• The servers dont have explicit knowledge of the
  clients resources
• Environment variables (load, network traffic
  ,memory etc )create a problem

5
Our Implementation

• made the entire system a resource aware system
• Designed a way for the server to keep a track of
  the client resources and dynamically adapt
  himself to changes in the resource status on the
  clients side.
• targeted those areas where resource utilization
  was a bottleneck and resource awareness would
  come in handy.

6
Experiment Environment
Dproc resource Monitoring
Exmd
Navigate Server
Visualization Client
Compression level
Filters
Visualization quality
7
Test Cases

• Case1 Network traffic monitoring
• Here the Number of packets are increased with
  time and frame rate is plotted

8

• Case 2 Load average monitoring
• Here the load average is increased with time
  and the frame rate is plotted

9

• Case3 Network Traffic monitoring
• Here the number of packets are increased with
  time and effect of detail factor is measured.

10

• Case4 Compression Levels
• This case is a specialized case of case 1 where
  we measured the effect of using different
  compression levels instead of just compress or
  not compress.
• Here the various compression levels do not
  really help performance.

11
Experiment Environment
Dproc resource Monitoring
Exmd
Navigate Server
Visualization Client
Compression level
Filters
Visualization quality
12
Environment Setup

• Syshack cluster
• PDSD Proactive Directory Service
• Echo Communication Channel
• Dproc Performance Monitoring Counter
• SmartPointer

13
(No Transcript)
14
Dproc Directory
15
(No Transcript)
16
Visualization of MD Data
17
Conclusion

• Environmental factors cause a bottleneck
• Resource monitoring is an efficient method to
  alleviate bottleneck
• Server dynamically adapts itself to suit the
  clients needs thus compensating for performance
  loss.

18
Future Work

• Complicated filter mechanism combine compression
  level and visualization quality
• Apply to multi-Server/multi-client scenario where
  complicated synchronization between server and
  client is needed.
• More environment information other than load
  average and network traffic be used

19
Reference

• Matt Wolf, Zhongtang Cai, Weiyun Huang, and
  Karsten Schwan, Smart Pointers Personalized
  Scientific Data Portals in Your Hand'',
  Supercomputing 2002, ACM, November 2002
• Resource-Aware Stream Management with the
  Customizable dproc Distributed Monitoring
  Mechanisms
• Greg Eisenhauer and Karsten Schwwan, The ECho
  Event Delivery System'', Technical Report
  GIT-CC-99-08, College of Computing, Georgia
  Institute of Technology, Atlanta, GA 30332-0280