Location Watergraafsmeer - IBM RS 6000/SP
('SOLO')
About SOLO
|
Access
|
The system can be accessed via the
Internet: the name of the interactive part
of the machine is solo.sp.sara.nl.
The recommended access tot the system
is via ssh and scp
The batch nodes are not accessible from
outside the SOLO system
|
|
Operating system
|
AIX (Unix)
|
|
Batch system
|
Loadleveller
|
|
Compilers
|
xlf: Fortran 77/90
xlc: C/C++
|
|
Libraries
|
- NAG: numerical and statistical
library
- IMSL: numerical and statistical
library
- ESSL: numerical library from
IBM
- P-ESSL: the parallel version of
ESSL
- SPRNG: Scalable Parallel Random
Number Generator
- the GNU suite
- DX: visualisation package
- SCALAPACK: parallel numerical
library
- PVM: parallel library
- MPI: parallel library
- POE: The Parallel Operating
Environment from IBM
|
The SP system consist of three "towers" or
"frames" and two separate servers. Two of the
frames contain eight so-called "nighthawk-2" nodes,
each of which contains 16 Power3 processors (375
MHz), with 1 Gbyte of memory per processor. The
third frame contains 3 "winterhawk-2" nodes, each
consisting of 4 Power3 processors with 1 Gbyte of
memory per processor. The two seperate servers are
p690 (also called "Regatta") servers, each
containing 32 Power4 processors (1.1 GHz), with 1
Gbyte of memory per processor.
The two nighthawk frames and the p690 servers
(128 + 64 processors respectively) will be used as
batch environment for scientific computations, the
winterhawk-2 frame will be used for special
applications.
The nodes of the SP system constitute a
so-called distributed SMP (symmetric
multiprocessor) system. Each node is an aggregate
of multiple processors (4 for winterhawk-2, 16 for
nighthawk-2 and 32 for regatta). Hence, within one
node the shared memory concept can be used in
programming, for which OpenMP is available. Of
course, MPI can be used as well for communication
between the processors. Between nodes you have to
use the distributed memory programming concept.
Communication between the nighthawk-2 nodes is
performed via the switch.
The main connection between the nighthawk-2
nodes is the so called "switch" (officially named
"SPswitch2"). The switch connects the nodes, not
the individual CPUs. Within a node the CPUs are
connected to an internal network that also
integrates 16 Gbytes of RAM-memory into a shared
memory addressing space.
A substantial part of the capacity is available
for Dutch researchers in general.
Location: SARA. Contact person: J.P. Hollenberg
(hollenberg@sara.nl).
About SARA
SARA Computing and Networking Services is a
center of expertise in the area of computers and
networks. SARA supplies a complete package of High
Performance Computing- and infrastructure services,
based on state-of-the-art information
technology.
SARA was founded in 1971 by the Free University
in Amsterdam, the University of Amsterdam and the
Mathematical Center. In 1985 SARA acquired the
status of National High Performance Computing
Center and hosted the Dutch national supercomputer.
Apart from this SARA hosts a number of other
supercomputers, some of which in co-ownership with
other universities.
SARA specializes in:
- High Performance Computing (production,
manipulation, simulation and visualization of
data)
- High Performance Networking
- Information Management
- Facility Management of computer systems and
networks.
SARA offers her customers:
- High performance computer systems: vector,
shared memory and distributed memory
- Experimental systems
- Virtual reality (CAVE and ImmersaDesk)
- Bulk data storage
- Advanced national and international
networks
- Vast expertise in the areas of High
Performance Computing and Networking,
parallelization (PECA), optimalization of
software, virtual reality and facility
management.
For further information please contact: SARA
Computing and Networking Services, Marketing and
Sales Department, P.O. Box 94613, 1090 GP
Amsterdam. Email: marketing@sara.nl.
Links
|