|
Department of Informatics Technische Universität München Informatik X: Rechnertechnik und Rechnerorganisation / Parallelrechnerarchitektur Prof. Dr. Arndt Bode , Prof. Dr. Hans Michael Gerndt |
|
|
|
Department of Informatics Technische Universität München Informatik X: Rechnertechnik und Rechnerorganisation / Parallelrechnerarchitektur Prof. Dr. Arndt Bode , Prof. Dr. Hans Michael Gerndt |
|
| Home | Addresses | Staff | Research | Lectures | ||
|
On top of the HAMSTER framework several high-level programming models have been implemented. In the following a short description of each of these models and their application area.
This programming model provides a fully transparent, distributed
thread packages based on the POSIX thread API, as defined in
the ANSI/IEEE Standard 1003.1. This creates an SMP-like environment
on top of a distributed cluster, enabling this architecture
to benefit from the large code basis of multithreaded applications.
The work on this model was initially started through the efforts
in the project by the
SISCI-Pthreads, which aimed at the establishment of a comprehenisive
software infrastructure based on standard APIs for SCI based clusters.
Similar to the distributed POSIX threads mentioned above, this
model provides a transparently distributed thread environment for
Windows NT (TM). It is based on Window's native thread API
included in the WIN32 (TM) programming environment.
The TreadMarks
(TM) system, developed and now commercially distributed by Rice University, is one of the most known
software DSM systems. Due to this, it is used by many people and research
groups and has a large code basis. This implementation of its API within
the HAMSTER framework, opens SCI clusters
to this code basis and allows these applications to directly benefit
from the underlying SAN and its HW-DSM capabilities.
The SPMD programming model provides a simple, yet efficient parallel
programming model allowing the parallel execution of programs across
the cluster in a synchronous manner. Especially resource allocations
have to be done synchronously by all nodes and synchronization is
mainly done using global barriers, even though also other
synchronization mechanisms are provided.
This implementation of the ANL macros (the subset used by the SPLASH-II
benchmark suite) is done directly on top of the native SCI-VM interface
allowing all SPLASH-II applications to directly run on the global
memory abstraction. As this benchmark is one of the most known, this
provides a powerfull evaluation instrument for the HAMSTER environemnt.
Questions, suggestions, ideas, criticism ? 
Distributed POSIX threads
Platform: Linux
Status: Only some minor functionlities missing
Distributed WIN32 (TM) threads
Platform: Win-NT
Status: Subset implemented
TreadMarks (TM) API
Platform: Win-NT and Linux
Status: Fully implemented
SPMD (Single-Program, Multiple-Data) model
Platform: Win-NT and Linux
Status: Fully implemented
ANL macros on top of the SCI-VM
Platform: Win-NT and Linux
Status: Subset necessary to run SPLASH-II implemented
Links to related topics in SMiLE
Back to SMiLE homepage
Don't hesitate to send us an
eMail!
| (none) | Webmaster |