GRAPE newletter vol.6 ( Nov 26 2004)

Dear Colleagues:

This is the sixth issue of our GRAPE newsletter, a brief summary of
recent developments regarding the GRAPE special-purpose computer for
stellar dynamics.  For further information: "http://www.astrogrape.org".

       +----------------------- CONTENTS:-----------------------------+
       | 1) The Next Grape: GRAPE-DR
       | 2) formation of a GRAPE Users Group
       | 3) GRAPEs Available During an N-body Summer School in Amsterdam
       | 4) Beowulf clusters using micro-GRAPE in Rochester and Heidelberg
       | 5) MOdeling DEnse STellar systems in Amsterdam
       | 6) Reports from the field: GRAPEs in New York
       +--------------------------------------------------------------+

1---------1---------1---------1---------1---------1---------1---------1

 THE NEXT GRAPE: GRAPE-DR

In May 2004, funding was announced for a new GRAPE project.
Unlike previous special-purpose computers in the GRAPE series,
the new GRAPE will be a programmable computer, containing
several hundred processors per chip.  The peak speed for N-body
simulations is expected to be 2 Petaflops, to be realized by 2008.
(see "http://grape.astron.s.u-tokyo.ac.jp/grape/computer/grape-dr.html").

The project is officially called "GRAPE-DR", with DR for "Data
Reservoir" (http://data-reservoir.adm.s.u-tokyo.ac.jp/).  Each
processor will contain one simple floating-point arithmetic unit and
integer arithmetic unit, and small local memory (256 double-precision
words).  All processors in one chip execute the same instruction in
SIMD fashion (as in the Connection Machine).  In fact, this new GRAPE,
which is also officially called GRAPE-DR, can be seen essentially as
a Connection Machine, but with the connections optimized for N-body
simulations.  A single chip will offer a peak speed of around 1 Tflops.
A single card (either PCI-Express or PCI-X) will house 4 chips, with a
peak speed of 4 Tflops, or around 30 times the current microGRAPE.

When used for N-body simulation, GRAPE-DR will "emulate" the present
GRAPE architecture, so that the programs which use GRAPE-6 will run
with little or no change.  GRAPE-DR will also be able to handle, for
example, SPH interactions between SPH particles, calculation of
multipole expansions in a FMM code, and possibly things like Linpack.

2---------2---------2---------2---------2---------2---------2---------2

 FORMATION OF A GRAPE USERS GROUP

We are pleased to announce the formation of a Users Group for the GRAPE 
family of computers. This will be an open email list, to foster discussion 
of the GRAPE hardware, installation issues, and the many aspects of GRAPE
programming. We hope to provide a pool of knowledge to help quickly resolve
common problems with the GRAPE, discuss programming techniques, and perhaps
even spur scientific collaborations. All generations of GRAPE architecture 
(and GRAPE users) are welcome.

To subscribe, send an email to the current list maintainer, Michael Sipior
(sipior@science.uva.nl), with "Grape Users Group subscription" in the
subject header.

3---------3---------3---------3---------3---------3---------3---------3

 GRAPES AVAILABLE DURING AN N-BODY SUMMER SCHOOL IN AMSTERDAM

From July 24 to July 30, 2005, an N-body Summer School will be held
in Amsterdam, the Netherlands.  Various computer facilities will be
available, including the local GRAPE-6 system as well as the CAVE
immersive visualization environment and the local parallel supercomputer.
For details see: http://carol.science.uva.nl/~spz/act/modest5c/

4---------4---------4---------4---------4---------4---------4---------4

 BEOWULF CLUSTERS USING MICRO-GRAPE IN ROCHESTER AND HEIDELBERG

The new micro-GRAPE cards have opened up a new and more flexible way
to use the power of GRAPE hardware in a Beowulf PC cluster environment.
Until recently one GRAPE6 ``dinosaur'' board of equivalent speed of 1
Tflops (if fully equipped; memory for 512k particles) was the smallest
unit of GRAPE to be connected to a host computer; it came in the form
of a special box with interfaces to be connected to the PCI interface
of a host computer.  Now a micro GRAPE6 offers 1/8 of the compute
power and (at the same equivalent compute speed) significantly more
memory for less than 1/8 of the price.  A micro GRAPE is compact enough
to be used as just another PCI card in a 4U PC box (but beware of the
rather large cooling unit, which may obstruct other slots).  This means
that we can build larger Beowulf PC clusters with integrated GRAPE
cards and a better balance between host and GRAPE capacities for
at least some applications (avoiding the host bottleneck, increasing
the number of communication channels between host and GRAPEs).  This
is interesting because it will allow a better use of more sophisticated
codes (e.g. codes based on Ahmad-Cohen neighbour scheme such as
NBODY6++) and SPH (smoothed particle hydrodynamics), in conjunction
with GRAPE clusters.  Two new installations of such clusters are
going into operation soon, one in Rochester (Rochester Institute of
Technology (RIT), New York State, USA, with David Merritt), where 8
nodes are already operational, and one in Heidelberg (Germany, with
A. Burkert, R. Maenner).   Both clusters will consist of 32 nodes with
32 micro-GRAPEs.

5---------5---------5---------5---------5---------5---------5---------5

 MODELING DENSE STELLAR SYSTEMS IN AMSTERDAM

On October 2nd, production was started on a 4-node GRAPE-6A/2 Beowulf
system called MoDeStA, for MOdeling DEnse STellar systems in Amsterdam,
the Netherlands.

The Beowulf is set up with 5 AMD 1.2GHz PCs with 100Mbit ethernet
interconnect and a 3Tbyte storage facility.  The GRAPE cluster is
hosted by the Dutch national supercomputer center SARA and financed by
the Dutch Organization of Science (NWO) and the University of
Amsterdam.  The official inauguration date will be announced on the
MoDeStA web page (http://carol.science.uva.nl/~spz/act/modesta/index.html).

Further information about the machine and access policies can be found
at this web page or by sending an email to spz@science.uva.nl

6---------6---------6---------6---------6---------6---------6---------6

 REPORTS FROM THE FIELD: GRAPES IN NEW YORK

 GRAPE6 Research at the American Museum of Natural History

 by Jarrod Hurley and Michael Shara (email: mshara@amnh.org)

The GRAPE program at AMNH began in 2000 when the long-term loan of a
16-chip prototype GRAPE-6 board was enabled by Jun Makino.  This
generous gift, in concert with the Stellar Collisions and Mergers
conference at AMNH in June 2000, and the arrival of Jarrod Hurley at
AMNH that year, jump-started the intensive star-cluster research now
underway in New York City. Generous donations from the Cordelia
Corporation, Hilary Lipsitz, Edward Norton and New York City funds
have lead to the subsequent purchase of five 32-chip GRAPE-6
boards. Some teething problems ...including balky power supplies...
were encountered, but after diligent work by our system administrator,
Dr. John Ouellette, we are now in the relatively luxurious position of
having five fully functioning 32-chip GRAPE-6 boards at our
disposal. Piet Hut's tireless efforts led to MODEST-1 being held at
AMNH, where Mike Shara hosted the LOC and could show off the new
Hayden Planetarium and its use as a GRAPE data display device.

Simulations performed to date on the GRAPE-6 boards at AMNH have 
primarily been driven by a desire to learn more about the 
evolution of star clusters - specifically to explore the nature 
of the stellar populations that reside in the star cluster 
environment and the effects of feedback between stellar, binary 
and cluster evolution. As such, being able to compare simulation 
results with observed data is essential. A big help in this 
respect is the Aarseth NBODY4 software which includes algorithms 
for dealing with stellar and binary evolution in concert with 
the gravitational calculations. This code was originally developed 
to interface with the GRAPE-4 board (circa 1996) so initial work 
focussed on adapting the interface to be consistent with the 
GRAPE-6 libraries (this was done with help from Sverre Aarseth and 
Jun Makino). After testing of the code we were ready (early 2001) to  
begin meaningful simulations. 

Not wanting to get too carried away with the additional computational
power offered by the GRAPE-6, we attempted rather modest initial
simulations, comprising N=20,000 stars with a 10% primordial binary
fraction. These open cluster models took approximately five days to a
week to complete (running on the prototype GRAPE-6 board).  Models
assuming a solar composition for the stars and models with the
metallicity reduced by a factor of five were performed. A remarkable
outcome of these simulations was the enhanced production rate of
short-period massive double-white dwarf binaries that we observed.
Relative to what we would have expected by evolving the same
populations outside of the cluster environment, i.e. with a population
synthesis code as opposed to NBODY4, we found that the merger rate of
these possible type Ia supernova progenitors was increased by a factor
of 10 owing to dynamical encounters. This interaction between a star
cluster and the stars it contains was also noted for other types of
exotic stars, such as blue stragglers, with many individual systems
taking wildly promiscuous evolution paths.

The arrival of two full GRAPE-6 boards allowed us to widen the scope
of our star cluster study and we began to look at the evolution of
binary-rich open clusters. Simulations starting with 12,000 single
stars and 8,000 binaries were performed and the large binary fraction
(40%) meant approximately three weeks was required on a 32-chip
GRAPE-6 board to evolve each model to completion (~8 Gyr). Motivated
by various observational studies, data from these simulations was used
to document the behaviour of the white dwarf population in a dense
star cluster.

We focussed on the white dwarf luminosity function and the white dwarf 
sequence in the colour-magnitude diagram, showing how these change with 
time and radial position as the cluster evolves. It was found that the 
cluster environment substantially increases the white dwarf mass fraction 
while also eroding the usefulness of white dwarfs as tracers of the 
initial mass function. To date we have completed a fairly large set of 
these binary-rich models and the aim is to carry out an extensive 
comparison of this data set with data available from observations of 
old open clusters, such as M67.
 
Now with five fully functioning GRAPE-6 boards at AMNH there is more
freedom to explore the star cluster parameter space. We are currently
looking at models of N=100,000 and N=200,000 stars with various
primordial binary fractions, up to 10% so far. Simulations of this
size are aimed at comparison with clusters such as M4 and NGC6397 for
which members of our team are involved with a number of Hubble Space
Telescope observing projects.

Observational searches for planets in the clusters M22 and 47 Tucanae
have also lead us to include planets in some of our simulations. An
early study involved taking ~2,000 of the stars in the N=20,000, 10%
binary simulations and placing a single Jupiter-mass planet in orbit
around each of these stars. This showed that a substantial population
of free-floating planets (for want of a better label) could be
generated via dynamical encounters between planetary systems and the
cluster stars and that this population could remain bound to the open
cluster on Gyr timescales. The survival prospects of planetary systems
in star clusters continues to be of interest to our group with
simulations proceeding in this area.

One final set of simulations to mention is a series of N=30,000 single
star models performed for a wide range of the metallicity assumed for
the cluster stars. Over the last three years or so much testing of the
GRAPE-6 boards has been required and if we wanted to test that a board
was functioning as normal, or perform a timing test, then we would
throw on one of these simulations which would typically take a day or
two to complete. They can be thought of as "engineering" or
"calibration" tests, exactly the sort of testing that regularly
happens on ground-based telescopes. Remarkably, they have proven to be
scientifically useful in that we were able to look at the effect of
metallicity (which changes the stellar evolution and mass-loss
timescales) on the internal dynamics of star clusters.

What's next? The evolution of multi-planet systems in star clusters,
and population studies in very dense star clusters rank as high
priorities.  Astrophysicists invariably push their tools to the limit,
and some of our current studies require 6-12 months of GRAPE6 time.
If only we could buy some GRAPE8 boards...

0---------0---------0---------0---------0---------0---------0---------0

           Piet Hut and Jun Makino
           (submissions to: piet@astrogrape.org or grape@astrogrape.org)


HOW TO (UN)SUBSCRIBE:  send an email message to grape@astrogrape.org
                       with Subject: (un)subscribe

00--------00--------00--------00--------00--------00--------00--------00