next previous
Next: BlockComm Up: Scalable Distributed Dynamic Grids: A Previous: Global Arrays (GA)

RSL

OVERVIEW: RSL [3] (Runtime System Library) (Argonne National Laboratory) is a Fortran library to support grid-based finite difference codes employing multiply nested grids. It is specifically targeted to weather forecasting codes and provides a Fortran 77 programming interface.

DYNAMIC/STATIC: Supports refinement of grids (domains); however am not sure if this can be done dynamically or has to be pre-defined

DIMENSIONS SUPPORTED: or 3 dimensions

DATA-TYPES SUPPORTED: Fortran 77 data-types

PROGRAMMING MODEL: RSL Objects: domain, message, stencil, broadcast/merges
Domain: computational grid
region of the domain can be refined to create nested domains
Message: list of variables that need to be communicated
message are specified by the compute node's code
Stencil: collection of messages to be exchanged (synchronously) with neighboring nodes
list of pre-defined stencils
Broadcast/Merge: intra-domain communication
list of messages that need to be communicated
user needs to supply packing/unpacking routines
RSL Process Structure: monitor node & compute nodes:
Monitor node:
specifies the domain(s)
specifies decomposition of domains
services request for I/O from compute nodes
Compute nodes:
defines communications objects
performs computation on local partitions of the domain

CONSISTENCY: Guaranteed for RSL stencil exchanges)

DISTRIBUTION: Blockwise decomposition - 2-D only

ACCESS SCHEME: Intra-grid communication:
Synchronous exchange communication based on defined stencil
Inter-grid communication:
Synchronous broadcast/merge communication

PROGRAMMING INTERFACE: Fortran 77

COMMUNICATION HARNESS: Chameleon on IBM SP-1
PICL on Intel Delta

PLATFORMS: SP-1, Delta

TARGET APPLICATION DOMAIN: Weather forecasting models (Penn State/NCAR Mesoscale Model Version 5)
Regular grid-based geophysical models

MISCELLANEOUS FEATURES: Input/Output support

TOOL HOME PAGE: RSL



next previous
Next: BlockComm Up: Scalable Distributed Dynamic Grids: A Previous: Global Arrays (GA)



Manish Parashar, parashar@cs.utexas.edu