Transcript Computing at the High Energy Physics Lab at FIT

Computing at the High Energy
Physics Lab at FIT
Patrick Ford, Jen Helsby, Richard
Hoch, David Pena
Dr. Hohlmann, Dr. Mitra
Current Projects
• Cluster Computing
- HEP’s computer cluster
• Grid Computing
- Getting the cluster on the Open Science Grid
• Simulations of Particles Through Matter
- Using Geant4 to model cosmic ray muons traveling through
different mediums
• Reconstruction Algorithms
- Developing algorithms to reconstruct muons passage through
matter
Cluster Computing
• Popular high performance computing solution.
• A computer cluster is a group of tightly coupled
computers that work together closely so that in
many respects they can be viewed as though they
are a single computer.
• Computing clusters make up over half of the top
500 most powerful computers in the world
System X at
Virginia Tech
(12.5
Teraflops)
[1]
HEP Computer Cluster
• Equipment loaned by University of Florida
• Started with 10 Dual CPU Intel Pentium 1.0
GHz servers
• One server for front-end, nine for nodes
• Uses networked attached storage (NAS)
• Cascaded switches for expandability and
redundancy
Cluster Topology
Current and Future Status
• Currently the original front-end is still being used,
but the cluster has expanded to 30 nodes
• Uses a high-end managed switch as the hub of
network and cascades to unmanaged switches with
10 nodes each
• Future expansion will include high-end compute
nodes, a ~10TB NAS, and a better front-end
MAGNUM
XV3045 NAS
[2]
HEP COMPUTER CLUSTER
Newest Nodes
NAS 1 and 2
Rocks
• Open-source Linux cluster distribution
• Enables end users to easily build
computational clusters
[3]
Rocks Kickstart Graph
Networked Attached Storage
• Also uses Rocks
• Uses RAID 5
- Faster writing. Each hard drive needs to write only 1/3 of
the data
- Efficiency increases as number of hard drives increases
- Fault tolerance. If any one hard drive fails, the data on
that drive can be reconstructed using the data from the
other two drives.
Condor
• Software that enables us to distribute the
load of a computing task over all the CPUs
in the cluster
• This type of software is called a batch job
system
• Well suited for grid computing, as it is able
to submit jobs to machines located all over
the world
Grid Computing
• A collection of networks, software, and computers
intended for shared use by organizations of people
• Resources are managed by a grid
• Users run applications as needed without worrying
about where the computers are
• Well-suited to organizations that consist of a large
number of geographically distributed members, all
working on a common project, and who require
shared computing resources in order to accomplish
their work
Grid Layers
I. Network layer
- Underlying connectivity
II. The grid's resources
- data storage, databases, software repositories, and even
sensors
III. The middleware, or "brains" of the grid
- does all the work to connect users' jobs to computing
resources
IV. Application layer
- diverse layer, as it includes virtually any program an end
user wishes to run
Open Science Grid
• A distributed computing infrastructure that is used
for large-scale scientific work
• Used by many universities, laboratories, and
software developers
• Backed by the NSF and the U.S. Department of
Energy's Office of Science
• The OSG Consortium builds and operates the
OSG project, with the goal of giving scientists
from many fields access to shared resources
worldwide
Science on OSG
• Scientists from many fields use OSG: particle and
nuclear physics, astrophysics, bioinformatics,
gravitational-wave science and computer science
collaborations
[4]
Getting On OSG
• Need the third layer, the middleware
• OSG’s is based on the Virtual Data Toolkit
(VDT)
• Installation package is needed, called
Pacman
• First installed Integration Test Bed (ITB)
client and then the Compute Element (CE)
package
Getting On OSG (cont.)
• Interfacing Globus and Condor
• Installing additional packages: Managed
Fork , MonaLisa, other monitoring services
• Getting personal and host certificates, and
the Certificate Authority (CA) list
• Testing and debugging the install
• Registering with OSG
Success… partially
The Integration Test Bed Map
Particle Simulations
• Geant4 provides a toolkit that enables
modeling of many different particles
through matter
• Much data can be extracted from these
simulations
• Our focus is on the simulation of cosmic ray
muons traveling through different mediums.
Why?
Muon Tomograpy
• Outgrowth of muon and proton radiography
• Provides a new way to detect threats such as
nuclear weapons or fissionable material,
and other terrorist threats (artillery shells,
IED’s, etc.)
• Why muons?
Why Muons?
• Relatively large elementary particles and travel at
relativistic speeds, can penetrate tens of meters into rocks
and other matter before attenuating as a result of
absorption or deflection by other atoms
• All natural occurring muons on Earth are due to cosmic
rays
• One per cm^2 per minute
• Muons are deflected by coulomb scattering, dependent on
the atomic number of the material
• Benefits over other techniques
Muons more penetrating
than gamma rays
No extra radiation dose
Fewer false alarms
[5]
How Does It Work?
Geant4
• Free tool that can run on Windows, Linux,
and MAC OS X
• Current version written in C++; former
versions written in Fortran
• Developed and maintained by the Geant4
collaboration which has over 100 members
worldwide
[6]
Scope of Geant4
• the geometry of the system (e.g. a box)
• the materials involved (e.g. Pb, U, etc.)
• the fundamental particles of interest (e.g.
electrons, muons, etc.)
• the physics processes governing particle
interactions
• the generation of event data
• the storage of events and tracks
• the visualization of the detector and particle
trajectories
• the capture and analysis of simulation data at
different levels of detail and refinement
First Scenario
• Created a 50x50x50 cm^3 lead block in an argon
atmosphere
• Bombarded it with 3GeV muons
• Interactions included:
muons, electrons - ionization, knock-on electrons,
multiple scattering
photons – absorption via photoelectric effect, Compton
scattering, pair production
Second Scenario
• Interfaced Cosmic RaY
(CRY) to simulate
cosmic ray muon
• Used same lead box as
material to detect, but
added detectors made of
G10 Material
 Blue – Positively Charged
Muons
Red – Negatively Charged
Muons
Green - Photons
Future Scenarios
• Adding more detectors
• Simulating a truck carrying plywood
• Hidden in the cargo area, and in the engine block
are small blocks of uranium
• Problems:
- Small amounts of high-z material harder to detect
- Engine block contains high-z material so multiple
scattering will occur
Reconstruction Algorithms
• Produces a 3D image from the projection
• For Muon Tomography two have been
prominently used:
- Point of Closest Approach
- Maximum Likelihood
• Implemented the POCA algorithm
Point of Closest Approach
• Take two lines, L1 and L2
• Take w(s,t) = L1(s) – L2(t)
• L1 and L2 closest when w(s,t) is a minimum, and
this vector is perpendicular to these points, meaning
w·v=0 and w·u=0
• Solve these by substituting w = L1(s)-L2(t) = w0 +
su - tv, where w0 = P0-Q0 we get two linear
equations:
• (u·u)s-(u·v)t=-u·w0 and (u·v)s-(v·v)t=-v·w0
• If we set a = (u·u), b = (u·v), c = (v·v), d = u·w0 and
e = v·wo and then solve for s and t we get the
equations:
• s = be-cd/(ac-b^2) and t = ae-bd/(ac-b^2)
Using POCA to find Scatter Points
3D Imaging
• Also have the scattering angle of the muon
• A large scattering angle indicates high Zobject
• When plotted, points will be assigned colors
according to scattering angle
[7]
Future Work
• Expanding the Cluster
- High-end nodes, NAS, and front-end
• Becoming a fully functional site on OSG
• Modeling more detailed scenarios using
Geant4
• Improving POCA, and implementing a
Maximum Likelihood algorithm
• Using real world data
References
1)
2)
3)
4)
http://www.vt.edu/spotlight/20060806_systemx.php
http://www.cybertronpc.com/
http://www.rocksclusters.org/wordpress/
http://www.opensciencegrid.org/Science_on_the_O
SG/Currently_Running_Applications
5) http://en.wikipedia.org/wiki/Muons
6) http://geant4.web.cern.ch/geant4/
7) http://math.lanl.gov/Research/Publications/Docs/bo
rozdin-2004-information.pdf