Maximum Performance Through Parallel Execution

By overcoming the traditional obstacles to parallel computing - costly hardware, programming difficulty, and lack of code portability - SCAI's Linda has become a powerful fulcrum for competitive advantage in many industries, from aerospace and automotive to petroleum, semiconductors, pharmaceuticals, and finance.

Linda provides a simple, yet complete command set which enables process creation, synchronization and communication. Every Linda software system employs powerful application optimization techniques and carefully tuned, architecture-specific run-time systems. Users benefit from, but are not concerned with, low-level details of different machine architectures or communication topologies.

The Linda Model

Linda is a coordination language which supplies the "glue" needed to cement many independent processes together into a single parallel program. Linda provides a virtual shared memory (VSM) that is logically shared by all the processes in a parallel program.

Processes in a Linda parallel program execute simultaneously and exchange data by generating, reading, and consuming data objects in the VSM using four basic operations:

• "out" places data into VSM
  
  
• "in" retrieves data from the VSM, simultaneously removing the corresponding data object
  
  
• "rd" copies data from the VSM without removing the corresponding data object
  
  
• "eval" used to create processes

Any program written in C or Fortran can be parallelized using these simple Linda operations.

Linda leaves to the base programming language the chores for which such languages were designed--arithmetic, loop control, procedure handling, and I/O -- while automatically and transparently handling the operations required to support many independent processes operating in parallel. As a result, parallel codes can often be deployed more rapidly and with less code modification, than with more time-consuming, low-level approaches to program conversion.

Linda Virtual Shared Memory Systems:

• Target homogeneous and heterogeneous networks of UNIX workstations, as well as shared or distributed-memory UNIX multiprocessors.
  
  
• Have extensive compile and runtime optimizations to maximize parallel performance gains
  
  
• Include easy-to-use tools for code development and debugging
  
  
• Are simple to deploy