Visualizing More Performance DataThan What Fits on Your Screen
Lucas Mello Schnorr(LIG – CNRS)
6th International Parallel Tools WorkshopStuttgart, Germany
September 25th, 2012
1/ 30
Challenges and MotivationSupercomputers today
– with very large applications
Sequoia (IBM BlueGene/Q) with 1,572,864 cores (#1 - Top500 - June/2012)
→ Sequoia (4 threads per core): 6 millions threads
Space/Time trace size explosion
Many entities in space + Very detailed behavior in time
Performance Visualization
→ How to keep the representation useful on scale?→ Considering the limited screen space available
2/ 30
Challenges and MotivationSupercomputers today – with very large applications
Sequoia (IBM BlueGene/Q) with 1,572,864 cores (#1 - Top500 - June/2012)
→ Sequoia (4 threads per core): 6 millions threads
Space/Time trace size explosion
Many entities in space + Very detailed behavior in time
Performance Visualization
→ How to keep the representation useful on scale?→ Considering the limited screen space available
2/ 30
Challenges and MotivationSupercomputers today – with very large applications
Sequoia (IBM BlueGene/Q) with 1,572,864 cores (#1 - Top500 - June/2012)
→ Sequoia (4 threads per core): 6 millions threads
Space/Time trace size explosion
Many entities in space + Very detailed behavior in time
Performance Visualization
→ How to keep the representation useful on scale?→ Considering the limited screen space available
2/ 30
Challenges and MotivationSupercomputers today – with very large applications
Sequoia (IBM BlueGene/Q) with 1,572,864 cores (#1 - Top500 - June/2012)
→ Sequoia (4 threads per core): 6 millions threads
Space/Time trace size explosion
Many entities in space + Very detailed behavior in time
Performance Visualization
→ How to keep the representation useful on scale?→ Considering the limited screen space available
2/ 30
Challenges and MotivationSupercomputers today – with very large applications
Sequoia (IBM BlueGene/Q) with 1,572,864 cores (#1 - Top500 - June/2012)
→ Sequoia (4 threads per core): 6 millions threads
Space/Time trace size explosion
Many entities in space + Very detailed behavior in time
Performance Visualization→ How to keep the representation useful on scale?→ Considering the limited screen space available
2/ 30
Trace visualizationReal BOINC availability trace file→ Availability is either true or falseOne volunteer client
GNUPlot to a vector file: 8-month and 12-day zoom
3/ 30
Trace visualizationReal BOINC availability trace file→ Availability is either true or falseOne volunteer clientGNUPlot to a vector file: 8-month and 12-day zoom
3/ 30
Trace visualization – trust the rendering?Acroread
Evince
Same vector file, two different views→ Should we trust the rendering?
4/ 30
Trace visualization – trust the rendering?Evince
Same vector file, two different views→ Should we trust the rendering?
4/ 30
Trace visualization – trust the rendering?Evince
Same vector file, two different views→ Should we trust the rendering?
4/ 30
Space/Time views
Widespread, useful, intuitive, fast adoptionAll trace events represented, causal order
Pajehttp://paje.sf.net
Vitehttp://vite.gforge.inria.fr
Vampirhttp://vampir.eu
However...
Also impacted by ever larger trace sizesLimited visualization scalability
5/ 30
Space/Time views
Widespread, useful, intuitive, fast adoptionAll trace events represented, causal order
Pajehttp://paje.sf.net
Vitehttp://vite.gforge.inria.fr
Vampirhttp://vampir.eu
However...
Also impacted by ever larger trace sizesLimited visualization scalability
5/ 30
Space/Time views – limitations
MPI Sweep3D – 16 processesSimulated with SMPI, traced with SimGrid→ pj_dump’ed to a csv file, loaded into RGantt-charts by R, dumped to a vector file
6/ 30
Space/Time views – limitationsEvince
6/ 30
Space/Time views – limitationsGhostscript
6/ 30
Space/Time views – limitations
MPI Sweep3D – 16 processesReal execution on the Griffon cluster of Grid’5000→ traced with TAU, converted with tau2paje
Once again, Gantt-charts by R→ vector file
7/ 30
Space/Time views – limitationsEvince
7/ 30
Space/Time views – limitationsAcroread
7/ 30
Space/Time views – closer look (ViTe tool)
Trust the OpenGL rendering, no data aggregation
Source: http://vite.gforge.inria.fr8/ 30
Space/Time views – closer look (new Pajé)Trust the rendering → without or with OpenGL
9/ 30
Space/Time views – closer look (new Pajé)Trust the rendering → without or with OpenGL
9/ 30
Space/Time views – closer look (old Pajé)
Opaque aggregating filter (no user interaction)→ Slashed rectangles represent time-integrated statesSelf-configure depending on temporal zoom
Source: http://paje.sourceforge.net10/ 30
Space/Time views – closer look (old Pajé)
Space dimension: one process per vertical pixel→ at best, 1000 process represented at the same time
11/ 30
Data aggregation for trace visualizationAlways present (in different forms)→ with large traces
Three groups
Implicit Data Aggregation→ uncontrolled, no visual feedbackExplicit Data Aggregation→ analyst control + visual feedbackForbidden Data Aggregation→ detect the amount of data, forbids overview
Tools might be classified in more than one group→ multiple approaches depending on data dimensionMain objective→ Explicit Data Aggregation
12/ 30
Data aggregation for trace visualizationAlways present (in different forms)→ with large traces
Three groups
Implicit Data Aggregation→ uncontrolled, no visual feedback
Explicit Data Aggregation→ analyst control + visual feedbackForbidden Data Aggregation→ detect the amount of data, forbids overview
Tools might be classified in more than one group→ multiple approaches depending on data dimensionMain objective→ Explicit Data Aggregation
12/ 30
Data aggregation for trace visualizationAlways present (in different forms)→ with large traces
Three groups
Implicit Data Aggregation→ uncontrolled, no visual feedbackExplicit Data Aggregation→ analyst control + visual feedback
Forbidden Data Aggregation→ detect the amount of data, forbids overview
Tools might be classified in more than one group→ multiple approaches depending on data dimensionMain objective→ Explicit Data Aggregation
12/ 30
Data aggregation for trace visualizationAlways present (in different forms)→ with large traces
Three groups
Implicit Data Aggregation→ uncontrolled, no visual feedbackExplicit Data Aggregation→ analyst control + visual feedbackForbidden Data Aggregation→ detect the amount of data, forbids overview
Tools might be classified in more than one group→ multiple approaches depending on data dimensionMain objective→ Explicit Data Aggregation
12/ 30
Data aggregation for trace visualizationAlways present (in different forms)→ with large traces
Three groups
Implicit Data Aggregation→ uncontrolled, no visual feedbackExplicit Data Aggregation→ analyst control + visual feedbackForbidden Data Aggregation→ detect the amount of data, forbids overview
Tools might be classified in more than one group→ multiple approaches depending on data dimension
Main objective→ Explicit Data Aggregation
12/ 30
Data aggregation for trace visualizationAlways present (in different forms)→ with large traces
Three groups
Implicit Data Aggregation→ uncontrolled, no visual feedbackExplicit Data Aggregation→ analyst control + visual feedbackForbidden Data Aggregation→ detect the amount of data, forbids overview
Tools might be classified in more than one group→ multiple approaches depending on data dimensionMain objective→ Explicit Data Aggregation
12/ 30
So, why controlled data aggregation?
Technical need: too much data to fit on small screens
Data aggregation→ key for large-scale visualizationSemantic: aggregated data→ more meaningful
13/ 30
So, why controlled data aggregation?
Technical need: too much data to fit on small screens
Data aggregation→ key for large-scale visualization
Semantic: aggregated data→ more meaningful
13/ 30
So, why controlled data aggregation?
Technical need: too much data to fit on small screens
Data aggregation→ key for large-scale visualizationSemantic: aggregated data→ more meaningful
13/ 30
So, why controlled data aggregation?
Technical need: too much data to fit on small screens
Data aggregation→ key for large-scale visualizationSemantic: aggregated data→ more meaningful
Goal: Visualization techniques for aggregated traces
13/ 30
Outline
1 Multi-Scale Data Aggregation
2 Visualization techniquesSquarified Treemap ViewHierarchical Graph View
3 Tools and framework
4 Conclusion
14/ 30
Multi-Scale Data Aggregation
15/ 30
Multi-Scale Data AggregationIn shortFΓ,∆ :R× T → R
(r , t) 7→∫∫
NΓ,∆(r ,t)ρ(r ′, t ′).dr ′.dt ′
time
space
Method requires dimension ordering→ Time scale is trivial – what about space scale?Additional scales could be added
Some Examples
time
space
time
space
time
space
16/ 30
Multi-Scale Data AggregationIn shortFΓ,∆ :R× T → R
(r , t) 7→∫∫
NΓ,∆(r ,t)ρ(r ′, t ′).dr ′.dt ′
time
space
Method requires dimension ordering→ Time scale is trivial – what about space scale?
Additional scales could be added
Some Examples
time
space
time
space
time
space
16/ 30
Multi-Scale Data AggregationIn shortFΓ,∆ :R× T → R
(r , t) 7→∫∫
NΓ,∆(r ,t)ρ(r ′, t ′).dr ′.dt ′
time
space
Method requires dimension ordering→ Time scale is trivial – what about space scale?Additional scales could be added
Some Examples
time
space
time
space
time
space
16/ 30
Multi-Scale Data AggregationIn shortFΓ,∆ :R× T → R
(r , t) 7→∫∫
NΓ,∆(r ,t)ρ(r ′, t ′).dr ′.dt ′
time
space
Method requires dimension ordering→ Time scale is trivial – what about space scale?Additional scales could be added
Some Examples
time
space
time
space
time
space
16/ 30
Multi-Scale Data Aggregation – Time
1 Time interval defined during the analysis2 Summary of events for each monitored entity
B
A
C
E
D
BlockedExecution
9 seconds
Time-integrated summary for processes
Numbers are in seconds (Execution, Blocked)B=(7,2)A=(4,5) C=(3,6) E=(4,5)D=(9,0)
17/ 30
Multi-Scale Data Aggregation – Time
1 Time interval defined during the analysis2 Summary of events for each monitored entity
B
A
C
E
D
BlockedExecution
9 seconds
Time-integrated summary for processes
Numbers are in seconds (Execution, Blocked)B=(7,2)A=(4,5) C=(3,6) E=(4,5)D=(9,0)
17/ 30
Multi-Scale Data Aggregation – Space1 Define a neighborhood for each monitored entity2 Apply an aggregating operator on the neighborhood
Neighborhood as a hierarchyResource-basedApplication groups
Deeper the hierarchy→ higher the quality
B
A
C
E
D
M1
M2
M3
C1
C2
G
Space-integrated summary
Aggregating operator: addition (Execution, Blocked)
B=(7,2)
A=(4,5)
C=(3,6)
E=(4,5)
D=(9,0)
M1=(11,7)
M2=(12,6)
M3=(4,5)
C1=(11,7)
C2=(16,11)
G=(27,18)
18/ 30
Multi-Scale Data Aggregation – Space1 Define a neighborhood for each monitored entity2 Apply an aggregating operator on the neighborhood
Neighborhood as a hierarchyResource-basedApplication groups
Deeper the hierarchy→ higher the quality
B
A
C
E
D
M1
M2
M3
C1
C2
G
Space-integrated summary
Aggregating operator: addition (Execution, Blocked)
B=(7,2)
A=(4,5)
C=(3,6)
E=(4,5)
D=(9,0)
M1=(11,7)
M2=(12,6)
M3=(4,5)
C1=(11,7)
C2=(16,11)
G=(27,18)
18/ 30
Visualization Techniques
19/ 30
Visualization techniques
Squarified Treemap ViewObserve outliers, differences of behaviorHierarchical aggregation
B Hierarchy: Site (10) - Cluster(10) - Machine (10) - Processor (100) C Hierarchy: Site (10) - Cluster(10) - Machine (10) - Processor (100) D Hierarchy: Site (10) - Cluster(10) - Machine (10) - Processor (100) E Maximum Aggregation
Hierarchical Graph ViewCorrelate application behavior to network topologyPin-point resource contention
All nodes Clusters Sites Fullaggregation
Host
20/ 30
Visualization techniques
Squarified Treemap ViewObserve outliers, differences of behaviorHierarchical aggregation
B Hierarchy: Site (10) - Cluster(10) - Machine (10) - Processor (100) C Hierarchy: Site (10) - Cluster(10) - Machine (10) - Processor (100) D Hierarchy: Site (10) - Cluster(10) - Machine (10) - Processor (100) E Maximum Aggregation
Hierarchical Graph ViewCorrelate application behavior to network topologyPin-point resource contention
All nodes Clusters Sites Fullaggregation
Host
20/ 30
Squarified Treemap View
Scalable representation for hierarchies→ better visualization scalability for large trees
Space-filling top-down recursive layout algorithmNode value→ space occupied in the screenSquarified version→ keeps rectangles ratio close to 1
T=6
M=3 N=2 O=1
A=1 B=1 C=1 D=1 E=1 F=1
T=6 M=3N=2
O=1
A=1
B=1
C=1
D=1
E=1
F=1
21/ 30
Squarified Treemap View
Scalable representation for hierarchies→ better visualization scalability for large treesSpace-filling top-down recursive layout algorithm
Node value→ space occupied in the screenSquarified version→ keeps rectangles ratio close to 1
T=6
M=3 N=2 O=1
A=1 B=1 C=1 D=1 E=1 F=1
T=6 M=3N=2
O=1
A=1
B=1
C=1
D=1
E=1
F=1
21/ 30
Squarified Treemap View
Scalable representation for hierarchies→ better visualization scalability for large treesSpace-filling top-down recursive layout algorithm
Node value→ space occupied in the screenSquarified version→ keeps rectangles ratio close to 1
T=6
M=3 N=2 O=1
A=1 B=1 C=1 D=1 E=1 F=1
T=6
M=3N=2
O=1
A=1
B=1
C=1
D=1
E=1
F=1
21/ 30
Squarified Treemap View
Scalable representation for hierarchies→ better visualization scalability for large treesSpace-filling top-down recursive layout algorithm
Node value→ space occupied in the screenSquarified version→ keeps rectangles ratio close to 1
T=6
M=3 N=2 O=1
A=1 B=1 C=1 D=1 E=1 F=1
T=6 M=3N=2
O=1
A=1
B=1
C=1
D=1
E=1
F=1
21/ 30
Squarified Treemap View
Scalable representation for hierarchies→ better visualization scalability for large treesSpace-filling top-down recursive layout algorithm
Node value→ space occupied in the screenSquarified version→ keeps rectangles ratio close to 1
T=6
M=3 N=2 O=1
A=1 B=1 C=1 D=1 E=1 F=1
T=6 M=3N=2
O=1
A=1
B=1
C=1
D=1
E=1
F=1
21/ 30
Squarified Treemap View – an exampleSynthetic trace with 100 thousand processesTwo states, four-level hierarchyVisualization artifacts without spatial aggregation
A Hierarchy: Site (10) - Cluster(10) - Machine (10) - Processor (100)
22/ 30
Squarified Treemap View – an exampleSynthetic trace with 100 thousand processesTwo states, four-level hierarchy
Visualization artifacts without spatial aggregation
B Hierarchy: Site (10) - Cluster(10) - Machine (10) - Processor (100)
22/ 30
Squarified Treemap View – an exampleSynthetic trace with 100 thousand processesTwo states, four-level hierarchy
Visualization artifacts without spatial aggregation
C Hierarchy: Site (10) - Cluster(10) - Machine (10) - Processor (100)
22/ 30
Squarified Treemap View – an exampleSynthetic trace with 100 thousand processesTwo states, four-level hierarchy
Visualization artifacts without spatial aggregation
D Hierarchy: Site (10) - Cluster(10) - Machine (10) - Processor (100)
22/ 30
Squarified Treemap View – an exampleSynthetic trace with 100 thousand processesTwo states, four-level hierarchy
Visualization artifacts without spatial aggregation
E Maximum Aggregation
22/ 30
Squarified Treemap View – KAAPI exampleKAAPI (run DAGs, work stealing for load balacing)188 processes running on five clusters
Rennes
Toulouse
Porto Alegre Bordeaux
Nancy
~43 s
~110 s~148 s
~78 s ~65 s
~67 s
~17 s
Analysis: stealing requests depends on latencyPorto Alegre – France: ~300 ms In France: ~10 ms
23/ 30
Squarified Treemap View – KAAPI exampleKAAPI (run DAGs, work stealing for load balacing)188 processes running on five clusters
Rennes
Toulouse
Porto Alegre Bordeaux
Nancy
~43 s
~110 s~148 s
~78 s ~65 s
~67 s
~17 s
Analysis: stealing requests depends on latencyPorto Alegre – France: ~300 ms In France: ~10 ms
23/ 30
Hierarchical Graph View
Scalable representation for graphsTopology, with application-level metricsIdentify resource bottleneck in space and time
Use spatial-temporal aggregated tracesInteractive force-directed layout (Barnes-Hut algorithm)
Trace metrics → geometrical properties→ Size, shape, filling, colors→ Nodes: monitored entities→ Edges: relationship among entities
hostA hostB
link
link utilization
time slice
24/ 30
Hierarchical Graph View
Scalable representation for graphsTopology, with application-level metricsIdentify resource bottleneck in space and time
Use spatial-temporal aggregated tracesInteractive force-directed layout (Barnes-Hut algorithm)
Trace metrics → geometrical properties→ Size, shape, filling, colors→ Nodes: monitored entities→ Edges: relationship among entities
hostA hostB
link
link utilization
time slice
24/ 30
Hierarchical Graph View
Scalable representation for graphsTopology, with application-level metricsIdentify resource bottleneck in space and time
Use spatial-temporal aggregated tracesInteractive force-directed layout (Barnes-Hut algorithm)
Trace metrics → geometrical properties→ Size, shape, filling, colors→ Nodes: monitored entities→ Edges: relationship among entities
hostA hostB
link
link utilization
time slice
24/ 30
Hierarchical Graph View - an example
Squares are hosts, diamonds are network linksColors represent different applicationsor parts of it (task type, phase)
Two clusters interconnected by four network links
time slice
25/ 30
Hierarchical Graph View - a larger exampleSquares are hosts, computer power defines sizeFrench Grid5000 platform: 2170 nodes
All nodes
26/ 30
Hierarchical Graph View - a larger exampleSquares are hosts, computer power defines sizeFrench Grid5000 platform: 2170 nodes
Clusters
26/ 30
Hierarchical Graph View - a larger exampleSquares are hosts, computer power defines sizeFrench Grid5000 platform: 2170 nodes
Sites
26/ 30
Hierarchical Graph View - a larger exampleSquares are hosts, computer power defines sizeFrench Grid5000 platform: 2170 nodes
Host
Fullaggregation
26/ 30
Tools and Framework
27/ 30
Open-source toolsPajeNG – Next Generation (Space/Time view), GPL3
http://github.com/schnorr/pajeng/
Only pj_dump and pj_validate (partial pajeng)Link against libpaje.so – then use Paje API to get data
Data Liberation Front®
$ ./pj_dump input.paje > output.csv
Then use your preferred visualization tool (gnuplot, R, GNU Octave, ...)
Viva (Aggregated Treemaps, Hierarchical graph), GPL3
http://github.com/schnorr/viva/
Partial graph visualization for now (soon)Deprecates Triva, serves as research framework
Some tracers and converters: SimGrid, Akypuera(with Score-P support, such as otf22paje), Poti,XKaapi, EZTrace, rastro2paje, libRastro, GTG,JRastro and counting...
$ ./otf22paje ./scorep-20120827/traces.otf2 | ./viva ...$ ./otf22paje ./scorep-20120827/traces.otf2 | pj_dump > output.csv
...
28/ 30
Open-source toolsPajeNG – Next Generation (Space/Time view), GPL3
http://github.com/schnorr/pajeng/
Only pj_dump and pj_validate (partial pajeng)Link against libpaje.so – then use Paje API to get dataData Liberation Front®
$ ./pj_dump input.paje > output.csv
Then use your preferred visualization tool (gnuplot, R, GNU Octave, ...)
Viva (Aggregated Treemaps, Hierarchical graph), GPL3
http://github.com/schnorr/viva/
Partial graph visualization for now (soon)Deprecates Triva, serves as research framework
Some tracers and converters: SimGrid, Akypuera(with Score-P support, such as otf22paje), Poti,XKaapi, EZTrace, rastro2paje, libRastro, GTG,JRastro and counting...
$ ./otf22paje ./scorep-20120827/traces.otf2 | ./viva ...$ ./otf22paje ./scorep-20120827/traces.otf2 | pj_dump > output.csv
...
28/ 30
Open-source toolsPajeNG – Next Generation (Space/Time view), GPL3
http://github.com/schnorr/pajeng/
Only pj_dump and pj_validate (partial pajeng)Link against libpaje.so – then use Paje API to get dataData Liberation Front®
$ ./pj_dump input.paje > output.csv
Then use your preferred visualization tool (gnuplot, R, GNU Octave, ...)
Viva (Aggregated Treemaps, Hierarchical graph), GPL3
http://github.com/schnorr/viva/
Partial graph visualization for now (soon)Deprecates Triva, serves as research framework
Some tracers and converters: SimGrid, Akypuera(with Score-P support, such as otf22paje), Poti,XKaapi, EZTrace, rastro2paje, libRastro, GTG,JRastro and counting...
$ ./otf22paje ./scorep-20120827/traces.otf2 | ./viva ...$ ./otf22paje ./scorep-20120827/traces.otf2 | pj_dump > output.csv
...
28/ 30
Open-source toolsPajeNG – Next Generation (Space/Time view), GPL3
http://github.com/schnorr/pajeng/
Only pj_dump and pj_validate (partial pajeng)Link against libpaje.so – then use Paje API to get dataData Liberation Front®
$ ./pj_dump input.paje > output.csv
Then use your preferred visualization tool (gnuplot, R, GNU Octave, ...)
Viva (Aggregated Treemaps, Hierarchical graph), GPL3
http://github.com/schnorr/viva/
Partial graph visualization for now (soon)Deprecates Triva, serves as research framework
Some tracers and converters: SimGrid, Akypuera(with Score-P support, such as otf22paje), Poti,XKaapi, EZTrace, rastro2paje, libRastro, GTG,JRastro and counting...
$ ./otf22paje ./scorep-20120827/traces.otf2 | ./viva ...$ ./otf22paje ./scorep-20120827/traces.otf2 | pj_dump > output.csv
...
28/ 30
Conclusion and Future WorkLarge scale traces – limited screen space
Explicit Data aggregation→ Controlled + visual feedback
Concerns with behavior attenuationAggregation may remove important detailsFlexible aggregation: operators & neighborhood
Visualization techniques for aggregated dataContinuous evaluation of visualization scalability
With larger data-sets, does it remain useful?
Future Work
Revisit Space/Time representationsAggregation operators to deal with time uncertaintyTo quantify loss of information when aggregating
29/ 30
Conclusion and Future WorkLarge scale traces – limited screen space
Explicit Data aggregation→ Controlled + visual feedback
Concerns with behavior attenuationAggregation may remove important detailsFlexible aggregation: operators & neighborhood
Visualization techniques for aggregated dataContinuous evaluation of visualization scalability
With larger data-sets, does it remain useful?
Future Work
Revisit Space/Time representationsAggregation operators to deal with time uncertaintyTo quantify loss of information when aggregating
29/ 30
Conclusion and Future WorkLarge scale traces – limited screen space
Explicit Data aggregation→ Controlled + visual feedback
Concerns with behavior attenuationAggregation may remove important detailsFlexible aggregation: operators & neighborhood
Visualization techniques for aggregated dataContinuous evaluation of visualization scalability
With larger data-sets, does it remain useful?
Future Work
Revisit Space/Time representationsAggregation operators to deal with time uncertaintyTo quantify loss of information when aggregating
29/ 30
Conclusion and Future WorkLarge scale traces – limited screen space
Explicit Data aggregation→ Controlled + visual feedback
Concerns with behavior attenuationAggregation may remove important detailsFlexible aggregation: operators & neighborhood
Visualization techniques for aggregated dataContinuous evaluation of visualization scalability
With larger data-sets, does it remain useful?
Future Work
Revisit Space/Time representationsAggregation operators to deal with time uncertaintyTo quantify loss of information when aggregating
29/ 30
Thank you for your attentionSome references
Visualizing More Performance Data Than What Fitson Your Screen. Lucas Mello Schnorr, ArnaudLegrand. The 6th International Parallel ToolsWorkshop. Springer. 2012. (Invited paper)Detection and Analysis of Resource Usage Anomalies in LargeDistributed Systems Through Multi-scale Visualization. Lucas MelloSchnorr, Arnaud Legrand, Jean-Marc Vincent. Concurrency andComputation: Practice and Experience. Wiley. 2012.A Hierarchical Aggregation Model to achieve Visualization Scalability inthe analysis of Parallel Applications. Lucas Mello Schnorr, GuillaumeHuard, Philippe Olivier Alexandre Navaux. Parallel Computing. Volume38, Issue 3, March 2012, Pages 91-110.
More information→ http://mescal.imag.fr/membres/lucas.schnorr/
INFRA-SONGS Project (WP-7)http://infra-songs.gforge.inria.fr/
Simulation of Next Generation SystemsWP-7: Visualization and Analysis
30/ 30