The CyberInfrastructure Day is a showcase of the modern computing infrastructure and its new applications.
Some of the most exciting research have become possible because of today's computing power.
It has been commonly known that simulations of new models need computation power. Simulating the heat dispensing of Space Shuttle launching helps design the launch pad and the shuttle body materials better. Some of the simulations can even serve real-time life-critical decisions such as whether or not to abort the mission when some components are malfunctioning in the space.
Simulations are useful beyond physics. The biomedical benefit from simulating the reactions of drugs and cells; the chemistry benefits from simulating the reactions of new composites, etc.
With more and more social behavioral models being invented, the application now does not limit to only nature sciences. Economics benefits from market simulation; even anti-terrorists missions benefit from high speed real-time analysis and simulation.
Another data-intensive example: by parsing the transcription of politician speeches over the years (a huge corpus, according to the presenter, would take 36 years to consume if one can read War and Peace twice in a day), one can clearly see the trend of topic shifting, or the consensus or conflicting opinions between the two political parties.
It's also important to see that education and research institutes are providing necessary infrastructure to facilitate such research. Some of the presentations showcase some optimization of the system performance that are inspiring to service providers like us.
Research and education are the seeds of future advances, and universities and research institutes are the most important places to enable such activities. Therefore, realizing the need of the up-coming world and preparing for it is crucial for any institute to remain relevant. CI day made me understand that in a meaningful way.
= My Notes ==
The following notes are for my own future reference. Other attendees may find it helpful if they missed something during the event.
== Foley ==
global warming
tele-commuting: we are doing work that doesn't need to be done any more probably because of habit out of industrial revoluation (social)
massive computation: cpu-intensive, data-intensive, visualization
3 key energy projecs
Cal-tech, Berkeley lab: sonar energy -> consumer
nuclear
PSU: energy efficiency for old building;
goal: in 5 years, cut the cost 20%
building viritual lab; via computation and visualization (running models and simulations)
models needs to be easy enough for people to make quick and accurate decisions in the field
design: to attract adaptors, with higher efficiency and cost
example of Mathematica solving a problem in 3 minutes (and explained in 3 pages in the manual), while it used to be a whole Ph.D. dissertation
== Henry ==
HCMP centers
Moore's rule: 4 years -> 10% of current
specialized skills set needed for maintanance
infrastructures: cooling, power, new building design
data storage requirement: 470TB per year so far; keeps growing
baseline to scheduling the use of service, and the license
consolidated customer service
portal service -> providing a familiar environment (Windows) to save customer's time in learning new environment
100x in code simliar to MatLb
Army combat vehicle surviability
Navy Costal Ocean Modeling
Air Force Kestrel (real time calculating aerodynamics)
chemstry field is growing fastest
general problems takes more time to design than to compute
== Biswas ==
AMES center
exploration system (ESMD)
modeling and simlulations
level of complexity is beyond the reach of strictly experiment or tehoretical methods
high throughput
mostly for enginerring problems, e.g. launch pad flame trench simulations; space shuttle extenral fuel tank redesign; vehicle assembling building safty assessment
high leadership:
high-resolution blogal ocean and sea-ice data synthesis; rotary wing aerodynmics; simulation merging black holes and predicting their grativational wave signatures
HA (mission critical; can't wait):
space shuttle debris transport analysis; crew risk assemsment on mission abort; high-fidelity simulations for hurricane prediction
visualization: because there still isn't any autoatmic method to easily interpretate the data; it often shows something un-considered or un-expected
old system: costly maintanaice; foot-pring; energy; return
high data storage requirement: data not understood today may be understood in the future
hyperwall (array of displays): useful for display, concurrently, several sets of related images
Utilization: different, conflicting goals of users and system provider: users want it to be as empty as possible; system provider wants it to be as full as possible
climate modeling (global warming)
computation steering (automatic parameters adjustment)
NASA grant
== Collins ==
National Cancer Institute
turn around time is crucial -- every 56 seconds a person in the US dies of cancer
algorithms change very quickly (almost weekly?!)
data-mining of old data that we didn't udnerstand
very much needed: highly-ksillked computational analysts in informatics
visuatlization
HCI (interface)
infrastructure:
biology is now digital; high data volume;
"last mile" problem: user; how to interpretate
mobile devices also now part of infrastructure
a few areas applicaation
microarrays and genotyping
sequencing technologies
imaging
nanobiomedcine
paradiam shifting in data storage: how long can we keep the data around? (It consums so much power just to have them on harddrive)
challenge: time to produce more data is also shorter
sequence -> treament: "time to solution" is crucial; patient info gets in, decision needs to be made as soon as possible
biological database http//biodbnet.abcc.ncifcrf.gov
Visualization:
real-time rendering of 3D images
leverage open source as much as possible;
nano-scale simulation and modeling
CSN
collaboration across ultiple displines required
shifting: researchers are shifting to manage their own research facility
question: optimization of human organization (architecture)
== Furlani ==
metrics is for measurement for improvement (both user experience and sys performance), within budget, and also good for proposal
ROI: resource utilization and metric; and grant database
Utilization Portal (Demand):
30-day utilization analysis to facilitate proper job queuing
open source (on SourceForge): UBMoD
TG:XD (?)
the staticstics include utilization, performance, and publication (ROI)
Colella's 7 Dwarfs (simulations in physical sciences)
biology has the most jobs, but usually short;
math; pysical sciences come second in # of jobs, but first in length
Grant Database
administration needs to evaluate the research result; how many resources are used and what are the results (how many publication; how many got awards)
(technical: irregular accumalative curves; how do we have hover-over popups with javascript? can we detect the color udnerneath? check the source.)
== Panel ==
=== Reel ===
Internet2: targeting at the research and education
National LambdaRail
regional dvp: many states have that but Pennsylvania is limited to geographic challenge; also coperation/collabtion between state agencies, public and private concerns
Penn State does its own, bought from the open market
there's capital funding available
48 strands; 1600+ route miles fiber
=== Huntoon ===
3ROX - middle mile
transparency
L3 tcp/ip
L2 circuit swtiched or ethernet (bandwidth on demand)
service: profiling of system (client) to find bottleneck
NAPD automaitc diagnostic tool to find bottleneck on the route
HPN-SSH: tuned for performance SSH
Web100: kernal patched
Weg10G: tuned for 10GE performance
network statistcis is useful in evaluating the distributed computation
Internet2 ION - dynamic routing
OpenFlow
iam.ps.edu
== Monroo Politician talks a lot ==
political science
politician spoken records are huge
statistical model for words mentioned
(compare words the two parties mention; color coded above and below the horizontal line)
see the trends
unified around 911; polarized during the Iraqi war
statistics on tweeter, txt msg, etc.
transcription is cleaned-up
Question: what about a model for prediction?
== Panel ==
audience:
recording TBs/week of EEG from animals
Leave a comment