Http://prophesy.cs.tamu.edu Xingfu Wu Xingfu Wu, Valerie Taylor Department of Computer Science Texas...
-
Upload
shona-ross -
Category
Documents
-
view
213 -
download
0
Transcript of Http://prophesy.cs.tamu.edu Xingfu Wu Xingfu Wu, Valerie Taylor Department of Computer Science Texas...
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, Valerie TaylorXingfu Wu, Valerie TaylorDepartment of Computer ScienceDepartment of Computer Science
Texas A&M UniversityTexas A&M University
Third Annual Workshop on OptIPuter Networing/Backplane ArchitectureThird Annual Workshop on OptIPuter Networing/Backplane Architecture
UCSD, January 25, 2005UCSD, January 25, 2005
Performance Analysis and Monitoring of Volatile
Xin
gfu
Wu
<
wu
xf@
cs
.ta
mu
.ed
u>
htt
p:/
/pro
ph
es
y.c
s.t
am
u.e
du
Network ConfigurationNetwork Configuration
OPT2(10.10.1.2)
OPT3(10.10.1.3)
OPT4(10.10.1.4)
Gigabit Switch(10.0.0.1)
Optiputer
Internetcs.tamu.edu
eth1:10.10.1.6 eth0: 128.194.142.24
OPT1(10.10.1.1)
Optical Network(LEARN)
CPU (dual Opteron 240 MHz)CPU (dual Opteron 240 MHz)
ApplicationApplication driverdriver
GPU (nVidia Quadro FX 3000)GPU (nVidia Quadro FX 3000)
256MB256MB
Video MemoryVideo Memory On-chip Cache MemoryOn-chip Cache Memory
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 Volatile on 5-nodesSnapshot of Volatile on 5-nodes
Xin
gfu
Wu
<
wu
xf@
cs
.ta
mu
.ed
u>
htt
p:/
/pro
ph
es
y.c
s.t
am
u.e
du
Three Components of VolatileThree Components of Volatile
Volvis: main programVolvis: main program RenderingRendering
Scalable displayingScalable displaying
User interactionUser interaction
Optistore: data serverOptistore: data server Assist visualization dataset handling such as data Assist visualization dataset handling such as data
management, processing, representation and transportmanagement, processing, representation and transport
Filter raw volumetric data and produce visual objectsFilter raw volumetric data and produce visual objects
Transfer Function Editor (tfUI)Transfer Function Editor (tfUI) User interface for transfer function selectionUser interface for transfer function selection
Xin
gfu
Wu
<
wu
xf@
cs
.ta
mu
.ed
u>
htt
p:/
/pro
ph
es
y.c
s.t
am
u.e
du
Different Access ScenariosDifferent Access Scenarios
Master Server
GigabitNetwork
VolvistfUI
Optistore
Client1 Volvis
Optistore
Client2
Volvis
Optistore
Client3
Volvis
Optistore
Client4Volvis
Optistore
Scenario 2Scenario 2
Scenario 1Scenario 1
Xin
gfu
Wu
<
wu
xf@
cs
.ta
mu
.ed
u>
htt
p:/
/pro
ph
es
y.c
s.t
am
u.e
du
Different Access ScenariosDifferent Access Scenarios
Master
GigabitNetwork
VolvistfUI
Optistore
Slave1 Volvis
Slave3
Volvis
Slave4Volvis
Slave2
Volvis
Scenario 3Scenario 3
Xin
gfu
Wu
<
wu
xf@
cs
.ta
mu
.ed
u>
htt
p:/
/pro
ph
es
y.c
s.t
am
u.e
du
Raw DatasetsRaw Datasets
Name Dimensions Size (KB)
Protein64x64x64 64x64x64 256
Fuel64x64x64 64x64x64 256
Furdave160x255x75 160x255x75 2989
Foot256x256x256 256x256x256 16384
Geo256x256x256 256x256x256 16384
Geo440x290x198 440x290x198 24673
Xin
gfu
Wu
<
wu
xf@
cs
.ta
mu
.ed
u>
htt
p:/
/pro
ph
es
y.c
s.t
am
u.e
du
Data access time Data access time
Data access time vs Processors (Scenario 1)
0
2
4
6
8
1+1 1+2 1+3 1+4
Number of Processors
Tim
e (s
ecs)
protein64x64x64
fuel64x64x64
foot256x256x256
geo256x256x256
geo440x290x198
furdave160x255x75
Data access time vs Processors (Scenario 2)
0
2
4
6
8
1+1 1+2 1+3 1+4
Number of Processors
Tim
e (s
ecs)
protein64x64x64
fuel64x64x64
foot256x256x256
geo256x256x256
geo440x290x198
furdave160x255x75
Data access time vs Processors (Scenario 3)
0
5
10
15
20
1+1 1+2 1+3 1+4
Number of Processors
Tim
e (
secs) protein64x64x64
fuel64x64x64
foot256x256x256
geo256x256x256
geo440x290x198
furdave160x255x75
Xin
gfu
Wu
<
wu
xf@
cs
.ta
mu
.ed
u>
htt
p:/
/pro
ph
es
y.c
s.t
am
u.e
du
Comparison of three scenarios Comparison of three scenarios
Data access time on 1+4 nodes
02468
101214161820
Scenario 1 Scenario 2 Scenario 3
Tim
e (s
ecs)
protein64x64x64
fuel64x64x64
foot256x256x256
geo256x256x256
geo440x290x198
furdave160x255x75
Xin
gfu
Wu
<
wu
xf@
cs
.ta
mu
.ed
u>
htt
p:/
/pro
ph
es
y.c
s.t
am
u.e
du
Comparison of three scenarios Comparison of three scenarios
Percentage of the data access time vs total startup time on 1+4 nodes
0
20
40
60
80
100
120
Scenario1
Scenario2
Scenario3
Per
cen
tag
e (%
) protein64x64x64
fuel64x64x64
foot256x256x256
geo256x256x256
geo440x290x198
furdave160x255x75
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 Performance MonitoringSnapshot of Performance Monitoring
Xin
gfu
Wu
<
wu
xf@
cs
.ta
mu
.ed
u>
htt
p:/
/pro
ph
es
y.c
s.t
am
u.e
du
Short-term PlansShort-term Plans
Port the instrumented Volatile to a large-Port the instrumented Volatile to a large-scale optiputer testbed for analysis scale optiputer testbed for analysis (3/2005)(3/2005)
Analyze the performance of JuxtaView and Analyze the performance of JuxtaView and LambdaRam applications (6/2005)LambdaRam applications (6/2005)
Where possible, develop models of data Where possible, develop models of data accesses for the different visualization accesses for the different visualization applications (9/2005)applications (9/2005)
Continue collaborating with Jason’s group Continue collaborating with Jason’s group about viz applications (12/2005)about viz applications (12/2005)
Xin
gfu
Wu
<
wu
xf@
cs
.ta
mu
.ed
u>
htt
p:/
/pro
ph
es
y.c
s.t
am
u.e
du
Long-term PlansLong-term PlansYear 3:Year 3:
Continue with analyzing Volatile on large-scale Continue with analyzing Volatile on large-scale OptIPuter testbedOptIPuter testbed
Work with Jason and Luc on redesigning VolatileWork with Jason and Luc on redesigning Volatile Start working on JuxtaView and LambdaRamStart working on JuxtaView and LambdaRam
Year 4:Year 4: Work with new Volatile, LambdaRam, JuxtaView Work with new Volatile, LambdaRam, JuxtaView
on large-scale OptIPuter testbedon large-scale OptIPuter testbed Where possible, develop models of data Where possible, develop models of data
accesses for different viz applicationsaccesses for different viz applicationsYear 5:Year 5:
Continue collaborating with Jason’s group about Continue collaborating with Jason’s group about viz applicationsviz applications
Integrate performance monitoring with software Integrate performance monitoring with software stack to aid in making performance directed stack to aid in making performance directed decisionsdecisions