|
Science Results from Grids Summaries
A number of scientists presented results from work they and their colleagues have done on grid resources. We summarize the presentations from nanomedicine, astrophysics, nuclear physics, applied mathematics and experimental particle physics.
Applied Mathematics
What is the minimum number of tickets you must buy to guarantee that you hold a winning ticket (given a certain number of games the teams must play)? Dubbed “The Football Pool Problem”, it’s a stumper. Forty years of effort on this problem have yielded clear values for matches up to five games, but for six games, the result was known only to be between 65 and 73. Jeff Linderoth of Lehigh University shared with us some “secrets” about gambling programs. Such problems can be easily formulated as Integer Programs (IPs), and IPs are quite often successfully solved with powerful commercial software packages such as CPLEX and XPRESS-MP. It turns out that commercial IP packages can solve ALMOST any IP – but not this one. In light of this, Linderoth described his team’s plan of attack -- it includes several tricks, massive computing power and a C++ software package called Master-Worker that encapsulates the abstractions of the standard Master-Worker paradigm. For six games, there are about 1000 IPs to run. So far these have run for about 200 CPU years on OSG resources, narrowing the range of the solution to between 71 and 73, inclusive.
Astrophysics
Steve Kent of Fermilab discussed Astrophysical Science with Grids. He emphasized that grid computing is an integral part of current operations and plans for future astrophysics experiments. At present, the Sloan Digital Sky Survey has accumulated color digital images of one quarter of the north sky and redshifts of one million galaxies and quasars, processed and stored on the grid. The proposed Dark Energy Survey, four times more powerful than SDSS, will accumulate similar data for the southern sky. Currently DES runs its simulations on the grid. DES' completed ImSim 2 data challenge used 15,000 CPU-hours and collected 3.4 TB; its planned ImSim 3 challenge will be five times larger.
We learned about types of astronomical data and how much information is associated with each type. SDSS matches resources to the job size: smaller jobs run on the local dedicated site, larger ones use FermiGrid resources, and still larger ones branch out to additional OSG sites. Kent described the “Galaxy Cluster Workflow” to illustrate the complexity and overlapping nature of astronomical data, showed an image of dark matter acting as a zoom lens, and presented SDSS Supernovae Survey first results.
Experimental Particle Physics
“Why build particle accelerators?” is the first question Ashutosh Kotwal of Duke University and CDF asked his audience. He then treated us to a synopsis of a century of particle physics research, theoretical and experimental, demonstrating how twelve fundamental matter particles fit neatly into an elegant mathematical framework, and how the framework is also used to predict the nature of fundamental forces. But then, the problem surfaced: this highly successful theory predicts that all particles should be massless. Theorists have postulated a “Higgs field” in order to address this, ahem, rather egregious inconsistency. Now it’s up to the experimentalists to find this field and its associated particle, the Higgs boson -- or it’s back to the drawing board.
To this end, CDF and the other big high energy physics experiments, present and future, are investing heavily in data simulation and reconstruction efforts. These efforts require sophisticated programs and large amounts of CPU. Since each collision event from an accelerator is independent of all others, events can be processed in parallel on different computers. CDF has clocked over a billion simulation events on OSG resources and D-Zero has reconstructed over 500 million events. Comparison of current experimental results to theory shows that if CDF and D-Zero analyze five times more data at Fermilab’s Tevatron, they have a good chance of discovering the Higgs boson. On the way to that goal, they’ve been zeroing in on the Higgs boson mass by making precision measurements of the top quark and W boson masses, confirming some key theoretical predictions, and searching for new fundamental symmetries of nature.
Kotwal seems to have answered not only the initial question, but also “Why build grids?”
Nanomedicine
From Reza Toghraee of UIUC we learned about the role ion channels play in heart pulsing, neuron and muscle cells, and toxins. He and his research colleagues are trying to model ion channels using a program called BioMOCA, based on Transport Monte Carlos (MOCA for MOnte CArlo), and look for ion traversal of a channel. This is a rare event, and therefore a large number of ion crossings must be detected in order to catch a single traversal. A recent application is the simulation of Mechanosensitive Channel of Small Conductance (MscS). They have run BioMOCA on both TeraGrid and OSG resources, and are expecting results within the next couple of months.
Nuclear Physics
Jerome Lauret of BNL took us back to the big bang and its resulting “quark gluon plasma,” and discussed how the STAR experiment studies this
plasma in recreated “little bang” conditions. Four experiments at BNL
are merging into two in the RHIC-II era, Phenix++ and STAR++. They will
confront a data rate 10 times that of their predecessors, and thus
larger event samples requiring more complex analysis.
The STAR Grid is currently focused on achieving its first
priority mission: making the massive amount of data available to its
participating sites in order to increase the analysis opportunities. The
STAR Grid consists of six main dedicated sites. They've recently tested on the non-dedicated FermiGrid site, as well, and plan to migrate most
simulation to OSG-based resources. STAR grid uses OSG infrastructure to
transfer data to remote sites, which makes data available to foreign
partners and allows balancing the analysis. Their transfer to China
late in 2006 increased the analysis 15%! STAR plans to run more than 80% of its Physics simulation on the OSG infrastructure by this summer.
|
