16
Xingfu Wu <[email protected]> h t t p : / / p r o p h e s y . c s . t a m u . e d u Xingfu Wu and Valerie Taylor Xingfu Wu and Valerie Taylor Department of Computer Science Department of Computer Science Texas A&M University Texas A&M University iGrid 2005, Calit2, UCSD, Sep. 29, 2005 iGrid 2005, Calit2, UCSD, Sep. 29, 2005 Performance Analysis and Monitoring of OptIPuter Visualization Applications

Performance Analysis and Monitoring of OptIPuter Visualization Applications

Embed Size (px)

DESCRIPTION

Performance Analysis and Monitoring of OptIPuter Visualization Applications. Xingfu Wu and Valerie Taylor Department of Computer Science Texas A&M University iGrid 2005, Calit2, UCSD, Sep. 29, 2005. Outline. Scalable Tiled Display System OptIPuter Viz Applications - PowerPoint PPT Presentation

Citation preview

Page 1: Performance Analysis and Monitoring of OptIPuter Visualization Applications

Xin

gfu

Wu

<

wu

xf@

cs

.ta

mu

.ed

u>

htt

p:/

/pro

ph

es

y.c

s.t

am

u.e

du

Xingfu Wu and Valerie TaylorXingfu Wu and Valerie Taylor

Department of Computer ScienceDepartment of Computer Science

Texas A&M UniversityTexas A&M University

iGrid 2005, Calit2, UCSD, Sep. 29, 2005iGrid 2005, Calit2, UCSD, Sep. 29, 2005

Performance Analysis and Monitoring of OptIPuter Visualization

Applications

Performance Analysis and Monitoring of OptIPuter Visualization

Applications

Page 2: Performance Analysis and Monitoring of OptIPuter Visualization Applications

Xin

gfu

Wu

<

wu

xf@

cs

.ta

mu

.ed

u>

htt

p:/

/pro

ph

es

y.c

s.t

am

u.e

du

OutlineOutline

Scalable Tiled Display SystemScalable Tiled Display System

OptIPuter Viz ApplicationsOptIPuter Viz Applications

Performance Analysis and MonitoringPerformance Analysis and Monitoring

Page 3: Performance Analysis and Monitoring of OptIPuter Visualization Applications

Xin

gfu

Wu

<

wu

xf@

cs

.ta

mu

.ed

u>

htt

p:/

/pro

ph

es

y.c

s.t

am

u.e

du

Scalable Tiled Display SystemScalable Tiled Display System

Take advantage of advanced graphics Take advantage of advanced graphics

pipeline technology and PC clusterspipeline technology and PC clusters Inexpensive, programmable GPUInexpensive, programmable GPU

Off-the-shelf, inexpensive PCs, high-speed networksOff-the-shelf, inexpensive PCs, high-speed networks

Tightly couple inexpensive LCD/TFT Tightly couple inexpensive LCD/TFT monitors to build a high-resolution display monitors to build a high-resolution display systemsystem

Replace the limited output resolution of standard devices Replace the limited output resolution of standard devices such as monitors, video projectors such as monitors, video projectorsFor detailed scientific visualizations with increased pixel For detailed scientific visualizations with increased pixel densitydensity

Page 4: Performance Analysis and Monitoring of OptIPuter Visualization Applications

Xin

gfu

Wu

<

wu

xf@

cs

.ta

mu

.ed

u>

htt

p:/

/pro

ph

es

y.c

s.t

am

u.e

du

OptIPuter Scalable Display Systems OptIPuter Scalable Display Systems

NCMIRSIO

UIC UIC

USGS EDC

TAMU

UCI SARA

UIUC/NCSA

Page 5: Performance Analysis and Monitoring of OptIPuter Visualization Applications

Xin

gfu

Wu

<

wu

xf@

cs

.ta

mu

.ed

u>

htt

p:/

/pro

ph

es

y.c

s.t

am

u.e

du

OutlineOutline

Scalable Tiled Display SystemScalable Tiled Display System

OptIPuter Viz ApplicationsOptIPuter Viz Applications

Performance Analysis and MonitoringPerformance Analysis and Monitoring

Page 6: Performance Analysis and Monitoring of OptIPuter Visualization Applications

Xin

gfu

Wu

<

wu

xf@

cs

.ta

mu

.ed

u>

htt

p:/

/pro

ph

es

y.c

s.t

am

u.e

du

OptIPuter Viz ApplicationsOptIPuter Viz Applications

Scalable Adaptive Graphics Environment Scalable Adaptive Graphics Environment (SAGE)(SAGE)

SAGE enabled Viz Tools:SAGE enabled Viz Tools: JuxtaView: JuxtaView: 2D high-resolution imagery2D high-resolution imagery Vol-a-Tile: Vol-a-Tile: 3D parallel volume 3D parallel volume

renderingrendering

Page 7: Performance Analysis and Monitoring of OptIPuter Visualization Applications

Xin

gfu

Wu

<

wu

xf@

cs

.ta

mu

.ed

u>

htt

p:/

/pro

ph

es

y.c

s.t

am

u.e

du

SAGESAGE

Manage the tiled displays as a virtual large Manage the tiled displays as a virtual large scale display screenscale display screen

Support different application executionsSupport different application executions Provide user-friendly interfacesProvide user-friendly interfaces

No reconfiguration need to relocate the No reconfiguration need to relocate the application windowsapplication windows

Freely re-arrange all application windows Freely re-arrange all application windows on the virtual screenon the virtual screen

Page 8: Performance Analysis and Monitoring of OptIPuter Visualization Applications

Xin

gfu

Wu

<

wu

xf@

cs

.ta

mu

.ed

u>

htt

p:/

/pro

ph

es

y.c

s.t

am

u.e

du SAGE Enabled Visualization Tool:

JuxtaViewSAGE Enabled Visualization Tool:

JuxtaView

JuxtaView – 2D high resolution image JuxtaView – 2D high resolution image montage displaymontage display

Panning and zooming of Panning and zooming of infinite resolution infinite resolution imagesimages..

Page 9: Performance Analysis and Monitoring of OptIPuter Visualization Applications

Xin

gfu

Wu

<

wu

xf@

cs

.ta

mu

.ed

u>

htt

p:/

/pro

ph

es

y.c

s.t

am

u.e

du

OutlineOutline

Scalable Tiled Display InfrastructureScalable Tiled Display Infrastructure

OptIPuter Viz ApplicationsOptIPuter Viz Applications

Performance Analysis and MonitoringPerformance Analysis and Monitoring

Page 10: Performance Analysis and Monitoring of OptIPuter Visualization Applications

Xin

gfu

Wu

<

wu

xf@

cs

.ta

mu

.ed

u>

htt

p:/

/pro

ph

es

y.c

s.t

am

u.e

du

OptIPuter Software ArchitectureOptIPuter Software Architecture

Optical Network Configuration

Novel Transport Protocols

Distributed Virtual Computer (Coordinated Network and Resource Configuration)

Visualization (SAGE, JuxtaView, Vol-a-Tile, …)

Applications (Neuroscience, Geophysics, …)

Perform

ance

Page 11: Performance Analysis and Monitoring of OptIPuter Visualization Applications

Xin

gfu

Wu

<

wu

xf@

cs

.ta

mu

.ed

u>

htt

p:/

/pro

ph

es

y.c

s.t

am

u.e

du

Performance MonitoringPerformance Monitoring

Graphics PerformanceGraphics Performance Rendering BW (MB/s): Amount of data Rendering BW (MB/s): Amount of data

being sent to display nodes being sent to display nodes Rendering rate (fps): Frames per second Rendering rate (fps): Frames per second

that rendering nodes are producingthat rendering nodes are producing Display BW (MB/s): Amount of data Display BW (MB/s): Amount of data

received by display nodesreceived by display nodes Display rate (fps): Frames per second that Display rate (fps): Frames per second that

display nodes are displayingdisplay nodes are displaying

Page 12: Performance Analysis and Monitoring of OptIPuter Visualization Applications

Xin

gfu

Wu

<

wu

xf@

cs

.ta

mu

.ed

u>

htt

p:/

/pro

ph

es

y.c

s.t

am

u.e

du

Graphics Performance by SAGEGraphics Performance by SAGE

Page 13: Performance Analysis and Monitoring of OptIPuter Visualization Applications

Xin

gfu

Wu

<

wu

xf@

cs

.ta

mu

.ed

u>

htt

p:/

/pro

ph

es

y.c

s.t

am

u.e

du

Performance MonitoringPerformance Monitoring

Monitor and explore the performance Monitor and explore the performance and behaviors of the applications on viz and behaviors of the applications on viz cluster in order to optimize themcluster in order to optimize them Machine loadMachine load CPU utilityCPU utility Memory usageMemory usage Network throughputNetwork throughput Disk usageDisk usage

Page 14: Performance Analysis and Monitoring of OptIPuter Visualization Applications

Xin

gfu

Wu

<

wu

xf@

cs

.ta

mu

.ed

u>

htt

p:/

/pro

ph

es

y.c

s.t

am

u.e

du

Snapshot of Cluster MonitoringSnapshot of Cluster Monitoring

Page 15: Performance Analysis and Monitoring of OptIPuter Visualization Applications

Xin

gfu

Wu

<

wu

xf@

cs

.ta

mu

.ed

u>

htt

p:/

/pro

ph

es

y.c

s.t

am

u.e

du

Performance Analysis Using ProphesyPerformance Analysis Using Prophesy

Code InstrumentationCode Instrumentation Measuring performance at different levelsMeasuring performance at different levels Identifying and locating performance Identifying and locating performance

bottlenecksbottlenecks Using kernel coupling techniques to explore Using kernel coupling techniques to explore

and quantify the interactions among the and quantify the interactions among the SAGE components / different applications in SAGE components / different applications in order to fine-tune SAGEorder to fine-tune SAGE

Performance comparison, modeling and Performance comparison, modeling and predictionprediction

Page 16: Performance Analysis and Monitoring of OptIPuter Visualization Applications

Xin

gfu

Wu

<

wu

xf@

cs

.ta

mu

.ed

u>

htt

p:/

/pro

ph

es

y.c

s.t

am

u.e

du Prophesy System

http://prophesy.cs.tamu.edu

Prophesy Systemhttp://prophesy.cs.tamu.edu

Web-based Prophesy GUI

Profiling & Instrumentation

Actual

Execution

Performance Database

TemplateDatabase

SystemsDatabase

ModelBuilder

SymbolicPredictor

DATACOLLECTION DATABASES

DATAANALYSIS