The document discusses high-throughput computing in engineering applications. It describes how most computing resources sit idle much of the time and how high-throughput computing can utilize these unused resources. Examples are provided of volunteer computing and distributed computing systems like Condor that can distribute engineering simulations and analyses across many computers. Case studies are presented on developing a seismic response database and performing probabilistic analysis of structural performance.

1. University
of Ljubljana
Faculty
of civil engineering
and geodesy
High-throughput computing in
engineering
CST & ECT 2008 Matevž Dolenc & Matjaž Dolšek
Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

2. Content
‣ Introduction
- Observations, utilisation of computing resources
‣ Computing environments
- Volunteer computing, high-throughput computing
‣ End-user scenarios
- A seismic response database
- A probabilistic performance assessment of a
structure
‣ Summary and future work
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

3. Important observations
‣ Modern research relies more and more on
computers - in-silica experiments.
‣ The majority of the world’s computing power is no
longer in supercomputer centres and institutional
machine rooms.
‣ Almost every organisation is sitting atop enormous
unused computing capacity that is widely distributed.
‣ In science and in engineering in general parametric
studies are of high importance.
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

4. Available computer resources
‣ UNIX® servers are actually “serving” something
less than 10 percent of the time.
‣ Most PCs do nothing for 95 percent of a typical day.
‣ Imagine ...
- ... an airline with 90 percent of its fleet on the
ground
- ... an auto maker with 40 percent of its assembly
plants idle
- ... a hotel chain with 95 percent of its rooms
unoccupied
Source: http://www-128.ibm.com/developerworks/grid/newto/index.html
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

5. Computing resource utilisation in computer
classrooms at UL-FGG
Source: http://grmada.fgg.uni-lj.si/condor-view
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

6. Computing resource utilisation in computer
classrooms at UL-FGG
Free Used
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

7. Computing resource utilisation in computer
classrooms at UL-FGG
Free Used Computer resource offline
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

8. Computing resource utilisation in computer
classrooms at UL-FGG
A computer classroom at UL-FGG
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

9. HPC vs HTC
‣ HPC – High Performance Computing
- Floating point operations per second (FLOPS)
- static environments, single large scale problems
- protocols: MPI, PVM, ...
‣ HTC – High Throughput Computing
- environment that can deliver large amounts of
processing capacity over long periods of time
- dynamic environments, parallel independent jobs,
parametric studies, ...
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

10. Volunteer computing
‣ Volunteer computing is a type of distributed
computing in which computer owners donate their
computing resources to one or more ”projects”
‣ Suitable for problems that are highly parallel in
nature, but not other types of parallel problems
‣ Example projects:
- SETI@Home, Einstein@Home, ...
‣ Different available software solutions:
- Berkeley Open Infrastructure for Network
Computing (BOINC)
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

11. Volunteer computing
‣ Technical issues
- heterogeneous distributed model, security,
connectivity, ...
‣ Non-technical issues
- Trust - can results be trusted / will project destroy
my data
‣ Why donate computing power?
- Help out science, help cure a disease, or any other
reason
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

12. Volunteer computing
Source: http://boinc.netsoft-online.com/e107 plugins/boinc/bp.php?project=2
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

13. High-throughput computing (HPC)
‣ HPC environment delivers large amounts of
computational power over a long period of time.
‣ Several computing system solutions:
- Condor, Torque, Sun Grid Engine ...
- Use of dedicated or non-dedicated computing
resources.
- Integration with different grid technology
software solutions
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

14. High-throughput computing (HPC)
‣ Condor
- ... is a batch queuing system to manage compute-
intensive jobs
- ... provides an application programming interface
compatible with Distributed Resource
Management Application (DRMAA)
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

15. High-throughput computing (HPC)
Source: http://www.cs.wisc.edu/condor/map/map.europe.color.large.gif
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

16. End-user scenarios - earthquake engineering
‣ A move from traditional single limit state
deterministic techniques to multiple performance
objectives in terms of probabilities of different
socio-economic decisions variables.
‣ A probabilistic performance assessment of a
structure
- difficult and time consuming task
- complex non-linear dynamic analyses
- # of analyses: different structural models (loss,
capacity, demand, seismic hazard), several ground
motion records, multiple levels of intensity
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

17. A seismic response database
‣ Develop tools to enable ?*2)/2.$)*-),4) !quot;#$%&'&(&)*)
?*2)/2.$4*)0,-)*$%&'&(&)*
reliable prediction of
seismic risk based on the
best knowledge regarding
earthquakes
‣ Seismic response database
6789:$)&2)/2.$4*)0,-)*$;)*4>*4$)2%*=
- new applications for E00<2.&'2,-)B
different end-users A 8-*<&)'2.$)0*.'4&
A F4*.*%*-.*$<2)'$,G$!quot;#
A #2)H$&))*))/*-'$,G$(4,&%$&4*&)
‣ Use of volunteer computing A$DDD
technology
‣ Application is currently
tested locally at UL-FGG
@-%A1)*4)B
A *-32-**4)
A 3,>*4/*-'
A 4*)*&4.C*4)
A$DDD
+,-&'*%$.,/01'2-3$4*),14.*)$5$
6789:$)*2)/2.$4*)0,-)*$;.<2*-'$)2%*=
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

18. A probabilistic performance assessment of a
structure
‣ A probabilistic performance assessment is de-aggregated into
independent models (loss, capacity, demand, seismic) ... IDA analysis
is used to determine the relation between the engineering demand
parameter and seismic intensity measure
‣ Each structural model has to be subjected to several ground motion
records, scaled to multiple levels of intensity in order to obtain
enough data about the behaviour of the structure
‣ Use of high-throughput computing technology
@87#%(quot;&$+7),#$(*9 AB#+CD<+,&'&6&-#-
!quot;quot;#$#%&'()*+%#quot;)%,-
!33$(quot;&'()*-E
F G3#*D##-
F H&'$&6
F CIquot;#$
F .)*,)%
;<!+&*&$=-(-+>)%?4$)>
.)*,)%+/01.2+3))$+)4+&5&($&6$#+quot;)738'(*9+
%#-)8%quot;#-+:+$)quot;&$+')+)%9&*(-&'()*
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

19. A probabilistic performance assessment of a
structure
‣ Performance analysis
- Number of analyses: 280
- Average analysis time: ~13 min
# of computers analysis time [hours] speed-up factor
1 61.3 1
5 14.7 4.17
10 7.1 8.63
25 2.5 24.52
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si

20. Summary and future work
‣ Different computing environments ware presented
- Enable various parametric studies or other highly
parallel analyses
- End-user scenarios and applications
- Use of spare computing resources
‣ Future work
- More research on volunteer computing is needed
- Integration with existing applications
- Offer better end-user experience
High-throughput computing in engineering CST & ECT 2008, Athens, Greece, 2-5 September 2008 mdolenc@itc.fgg.uni-lj.si