Astronomy

wiglaf

Astronomy

jhallum

This cluster is a multidepartment resource. The majority of the time it is used for tightly parallel problems such as hydrodynamic situations, but it is flexible enough to be used in loosely parallel problems. There are 6 different research groups using this cluster in 3 different departments (Astronomy, Math, and Physics), so it is difficult to cover all of the problems being worked on in this short space.

2002

2006-2007

16

32

0.6

Linux

Myrinet

Astronomy

(to be decided)

see notes

jhallum

???

2005

???

48

96

???

Linux

???

Contingent upon Drs. Nuria Calvet and Lee Hartmann accepting positions with the Department.

Biophysics Research Division

mercury / hermes

skrimm

dsglaser

???

2000

???

17

17 desktop machines

Linux

100 Mbps Ethernet

Chemistry

alpha

andricio

raeker

Our research explores theoretical and computational topics at the interface between structural molecular biology and physical chemistry. It hinges on the central theme of developing and applying computer and modeling methods to describe, in terms of dynamics and thermodynamics, biologically important molecular processes, with the aim to complement, enhance or predict experimental findings. One example is the /Computer Simulation of DNA-Binding Machines/. Protein DNA interactions are essential in such crucial cellular functions as replication, repair, transcription or recombination. Many enzymes at and ahead of the replication fork affect large DNA fragments. For instance, topoisomerases undo DNA knotting. Others, like helicases and polymerases, are biomolecular motors: they use the energy of binding and/or hydrolysis of nucleotides to do mechanical work on the DNA fragments to which they bind. We have an avid interest in the theoretical description of these fundamental genetic processes through massively parallel computer simulations. Together with postdocs and graduate students, we involve 5-6 undergraduates in such simulations each year.

2003

"until last node dies"

9

18

9 towers in one rack

Linux

GbE

Frontend has 560 GB data storage; nodes have 100 GB data storage and 2 GB RAM each. May move to Bioinformatics machine room.

Chemistry

thermis

nwalter

raeker

We use our 17-node dual-processor PC cluster largely for molecular dynamics (MD) simulations of large RNA enzymes in explicit water (about 9,000 atoms each). We use the software package AMBER 7.0 for the simulations and later analyze our time trajectories using scripts written in PERL. We are particularly interested in correlating the observed structural dynamics of the RNA enzyme in silico with experimental observations of such dynamics as observed by single molecule fluorescence spectroscopy. 1 nanosecond of MD simulation takes about 1 week calculation time on one node, and we typically like to repeat the same simulation several times as well as run a number of closely related simulations in parallel.

2003

"until last node dies"

17

34

17 towers in two racks

Linux

GbE

Frontend has 400 GB data storage; nodes have 80 GB storage and 2 GB RAM each.

Chemistry

hector

rrsharp

raeker

Our 16-node Beowulf cluster is used primarily for the calculation of NMR paramagnetic relaxation rates using Spin Dynamics Simulation techniques. The SD algorithms are similar in spirit to Molecular Dynamics simulations but involve quantum mechanical propagation of the electron spin operators driven by a stochastic hamiltonian rather than, in the= case of MD, classical propagation of molecular degrees of freedom. This capability is essential to our research and is used on a day-by-day basis.

2002

"until last node dies"

16

15 towers in one rack

Linux

GbE

100 GB disk space on frontend; 2 GB RAM per node

Chemistry

medussa

eitan

raeker

At the present time, we mostly use the cluster for running imaginary time path-integral and real-time classical molecular dynamics simulations of liquids, as well as Langevin and Brownian simulations of polymers. Such simulations require a lot of averaging, and having many CPU’s allows us to do this in a timely fashion. Those applications are also relatively light on memory, which significantly lowers the price per CPU. We do use MPI in order to “manage” the averaging in an efficient manner. More recently, we have become interested in applications that require multiple diagonalizations of very large matrices, which are memory intensive, and it is likely that our need for memory will rapidly grow in the future.

2002 (with nodes added periodically since)

"until last node dies"

51

51 towers in five racks

Linux

100 Mbps Ethernet

Frontend has 80 GB disk space 1 GB RAM; nodes have 40 GB and 1 GB RAM each.

Chemistry

summer

bdunietz

raeker

Ab initio electronic structure calculations

2004

"until last node dies"

20

40

1.5

Linux

GbE

All nodes have 4 GB RAM and 250 GB disk space. 5 nodes have additional 500 GB disk space.

Chemistry

zeus

lbartell

raeker

Zeus is dedicated to molecular dynamics simulations of the spontaneous homogeneous nucleation of crystals in supercooled liquids. The study of kinetics in this area of research has heretofore been notoriously difficult, whether by laboratory experiment or the application of theory, although many conjectures about what goes on during freezing have been proposed. Since the topic is of importance in technology and is a challenging topic in its own right in pure science, it is felicitous that computer simulations have finally begun to provide answers. These simulations are massively computer-intensive, however, and Zeus in its present configuration is barely adequate.

2002

"until last node dies"

48

48 towers in three racks

Linux

100 Mbps Ethernet

Frontend has 500 GB data storage; each node has 40 GB disk space and 1 GB RAM each. 16 more nodes might be added by end of 2005.

Chemistry

tiamat

Chemistry

raeker

Shared amoung various research groups who do not have their own cluster. General molecular mechanics simulations, ab initio electronic structure and crystalligraphic calculations.

2000

"until last node dies"

24

24 towers in two racks

Linux

GbE

Economics

(no name)

Economics Lab

aclrk

Our cluster is an xgrid system that is used by some grad students to perform Stata, Matlab and perl calculations in parallel. We use desktop lab OS X machines as cluster nodes.

2004

???

5

10

10 towers

MacOS X

building Ethernet

Geology

condor

poulsen

messinam

The Earth's climate record is one of global change. For example, in the last 3 million years, Earth's climate has fluctuated between glacial and interglacial states. The cause of these fluctuations is largely unknown. We are using three-dimensional models of the ocean and the atmosphere to understand the forcings and processes that have shaped Earth's climate history. Our climate code has been optimized for cluster application, allowing us to run long equilibrium simulations.

2004

???

17

34

1

Linux

Myrinet and GbE

This cluster will be merged into petrarch. 1.5TB of RAID storage served from master (master will become dedicated file server after merge with Petrarch)

Geology

petrarch

tehlers

messinam

The primary application of the cluster is in the simulation of the thermal, mechanical, and topographic evolution of tectonically active mountain ranges. Various 3D finite element models are coupled together to simulate the cooling history of rocks as a function of different tectonic settings and erosional histories. Model results are tested by comparison to geochemical data collected from the mountain range being simulated. Hundreds to thousands of simulations are typically conducted using an intelligent inverse scheme to find the range of model parameters (and hence geologic processes) that produce results which satisfy the data. Although most of the group's inverse model applications use an 'embarrassing parallel' approach to run a suite of simulations some applications do take advantage of the message passing interface (MPI).

2004-2005

???

33

50

2

Linux

GbE; 16 of the nodes are on Myrinet for poulsen

2.0TB of RAID storage served from a dedicated linux file server

Geology

trans

keken

messinam

The trans cluster is used for finite element modeling of mantle convection with emphasis on the long term chemical and physical evolution of the Earth's interior and the dynamics of mantle plumes and subduction zones; massively parallel rendering of seismic wave propagation through the Earth's interior; and development of parallel finite element modeling tools.

2003

???

37

2

Linux

GbE

2.0TB of RAID storage served from a dedicated linux file server

Geology

udocluster

ubecker

messinam

Mineral surface reactions are important in a number of environmental processes such as adsorption of organic and inorganica material, redox reactions, crystal growth and dissolution, the adsorption/incorporation of nuclear material, to name a few. In order to understand these processes at the molecular level and as a tool to understand molecular scale microscopy and spectroscopy methods, we perform quantum mechanical and, for larger systems, empirical force-field type calculations. These are done on a number of minerals, surfaces, adsorption mechanisms with coupled redox processes, and reactions involving organic species such as growth inhibitors, siderophores, pesticides, and polypeptides. Especially for quantum mechanical calculations, calculations can become very demanding. Programs such as Castep, VASP, NWCHEM, Gaussian, Crystal etc. can sometimes run for weeks on a parallel computer, just for a system containing several tens of atoms. Therefore, we are using a linux cluster with currently 20 nodes to accomodate these calculations.

2004

???

9

10

0.5

Linux

GbE

Geology

udo2

ubecker

messinam

Mineral surface reactions are important in a number of environmental processes such as adsorption of organic and inorganica material, redox reactions, crystal growth and dissolution, the adsorption/incorporation of nuclear material, to name a few. In order to understand these processes at the molecular level and as a tool to understand molecular scale microscopy and spectroscopy methods, we perform quantum mechanical and, for larger systems, empirical force-field type calculations. These are done on a number of minerals, surfaces, adsorption mechanisms with coupled redox processes, and reactions involving organic species such as growth inhibitors, siderophores, pesticides, and polypeptides. Especially for quantum mechanical calculations, calculations can become very demanding. Programs such as Castep, VASP, NWCHEM, Gaussian, Crystal etc. can sometimes run for weeks on a parallel computer, just for a system containing several tens of atoms. Therefore, we are using a linux cluster with currently 20 nodes to accomodate these calculations.

2005

???

12

0.5

Linux

GbE

LSA Information Technology

obsidian

LSAIT

kpachla, lyzhang

College-wide cluster for general faculty use. Intended to allow faculty to run small jobs, and to developing code to run on larger clusters at CAC or other supercomputing centers.

2005

2008

17

34

1

Linux

Infiniband, GbE

On order.

LSA Information Technology

lsa-cluster2

LSAIT

dpugh

College-wide cluster for general faculty use. Intended to allow faculty to run small jobs, and to developing code to run on larger clusters at CAC or other supercomputing centers.

2005

2008

17

34

1

MacOS X

not decided yet

To be ordered summer 2005, budget permitting.

LSA Information Technology

(to be decided)

LSAIT

kpachla, lyzhang

Internal LSAIT testing and development

2005

2009

3

6

0.15

Linux

not decided yet

To be ordered summer 2005, budget permitting.

LSA Information Technology

lsa-cluster1

LSAIT

dpugh

Internal LSAIT testing and development

2004

2008

5

10

0.25

MacOS X

GbE

Currently also being used by EEB faculty.

Physics

hypnos

ATLAS group (DoE), MGRID

smckee

Support the Michigan ATLAS groups computing needs as well as for some funded research programs (UltraLight, TeraPaths and GridNFS). The ATLAS work involves significant grid middleware which is provided by Grid3/Open Science Grid. Most simulations, reconstruction, digitization and analysis codes involved are "embarrasingly" parallel. Each code runs on a single processor with a minimum of 1 GB of RAM per processor.

2003

2007

14

27

1 + 7 towers

Linux

GbE

Utilized for both ATLAS and networking research, including 10 Gigabit ethernet, MPLS/QoS and GridNFS (based upon NFSv4). Storage requirements are significant. The group currently has about 12 Terabytes of storage in RAID5 arrays as well as scattered storage on single disks totaling about 22 Terabytes. Network connectivity also critical. Currently has access directly to campus gigabit backbone and MiLR 10 gigabit wavelength to Chicago/StarLight.

Physics

um-atlas

ATLAS group (DoE), MGRID

smckee

Support the Michigan ATLAS groups computing needs as well as for some funded research programs (UltraLight, TeraPaths and GridNFS). The ATLAS work involves significant grid middleware which is provided by Grid3/Open Science Grid. Most simulations, reconstruction, digitization and analysis codes involved are "embarrasingly" parallel. Each code runs on a single processor with a minimum of 1 GB of RAM per processor.

2003

2007

128

256

5

Linux

dual GbE

Utilized for both ATLAS and networking research, including 10 Gigabit ethernet, MPLS/QoS and GridNFS (based upon NFSv4). Storage requirements are significant. The group currently has about 12 Terabytes of storage in RAID5 arrays as well as scattered storage on single disks totaling about 22 Terabytes. Network connectivity also critical. Currently has access directly to campus gigabit backbone and MiLR 10 gigabit wavelength to Chicago/StarLight.

Physics

opus

evrard

mbusha

???

2004

no plan

34

68

1

Linux

GbE

* note: total number of racks does not count the tower systems

Mean:

25.95652174

41.91304348

Median:

17

34