kristopher blair jason duo li daniel moberly michael sedillo profiling performance of a web proxy on...

10
Kristopher Blair Jason Duo Li Daniel Moberly Michael Sedillo Profiling performance of a Web Proxy on a multi- core platform

Upload: paul-lambert

Post on 18-Jan-2016

212 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Kristopher Blair Jason Duo Li Daniel Moberly Michael Sedillo Profiling performance of a Web Proxy on a multi-core platform

Kristopher BlairJason Duo Li

Daniel MoberlyMichael Sedillo

Profiling performance of a Web Proxy on a multi-core platform

Profiling performance of a Web Proxy on a multi-core platform

Page 2: Kristopher Blair Jason Duo Li Daniel Moberly Michael Sedillo Profiling performance of a Web Proxy on a multi-core platform

Focusing on speed of web proxies for use in mobile networks – extremely common in modern age

File size is important in mobile networks – small devices and slower connections, therefore in need of smaller delays

File size vs. computation time tradeoffs are very important

Mobile proxies trans-code large images to small images for use on mobile devices – need to optimize the time it takes

Introduction/Motivation

Page 3: Kristopher Blair Jason Duo Li Daniel Moberly Michael Sedillo Profiling performance of a Web Proxy on a multi-core platform

Web proxy in iMPACT’s mobile computing lab – written to optimize JPEG image delivery

Image trans-coding: web proxy trans-codes JPEG images for use on mobile devices

Want to analyze bottlenecks to see how the speed of the proxy can be improved

Make sure proxy functions within acceptable levels of power and thermal output

Introduction/Motivation

Page 4: Kristopher Blair Jason Duo Li Daniel Moberly Michael Sedillo Profiling performance of a Web Proxy on a multi-core platform

Verify previous findings of the iMPACT research team

Evaluate performance versus power consumption and thermal outputs

Verify that the current web proxy design functions within acceptable levels of power consumption and thermal output.

Our Goals:

Page 5: Kristopher Blair Jason Duo Li Daniel Moberly Michael Sedillo Profiling performance of a Web Proxy on a multi-core platform

Approach

Our network layout

Page 6: Kristopher Blair Jason Duo Li Daniel Moberly Michael Sedillo Profiling performance of a Web Proxy on a multi-core platform

Approach Timestamps for each event on each machine

allows for timing analysis

Linux Lm_sensors for thermal data

“Watts Up?” meter for power consumption data

Page 7: Kristopher Blair Jason Duo Li Daniel Moberly Michael Sedillo Profiling performance of a Web Proxy on a multi-core platform

Earlier work shows that network overhead is significant, especially in smaller files – TCP overhead is quite significant

Larger files still show overhead in connection setup

Expected Results:

Results of previous analysisOn small files (300B – 70KB)

Page 8: Kristopher Blair Jason Duo Li Daniel Moberly Michael Sedillo Profiling performance of a Web Proxy on a multi-core platform

Expect thermal output and power consumption to fall along standard curves

Expected Results

Expected thermal output (°F) vs. Workload (Request /s)

Expected power consumption (Watts) vs. Workload (Request/s)

Page 9: Kristopher Blair Jason Duo Li Daniel Moberly Michael Sedillo Profiling performance of a Web Proxy on a multi-core platform

Presentation, Phase I, II & III report

lm_sensorConfiguration

Watts up

Proxy Coding Modification

Data Collection

Data Analysis

Dan Duo Kris Dan Dan Dan

Duo Michael Duo Duo Duo Duo

Michael Michael Michael

Kris Kris Kris

Distribution of Work

Page 10: Kristopher Blair Jason Duo Li Daniel Moberly Michael Sedillo Profiling performance of a Web Proxy on a multi-core platform

Questions?