Motivation

• Parallel Processing (multiprocessors, multicomputers, ...):
  high performance at reasonable cost
  (multiple units built with standard technology)

• Currently in market, often already present at client location:
  • multiprocessor workstations
    (2-4 Procs. - e.g., SPARC 20, multiprocessing PC's, ...)
  • group servers
    (4-8 Procs. - e.g., SPARC 1000, multiprocessing PC's, ...)
  • departmental servers
    (10-30 Procs. - e.g., SPARC 2000, SGI Challenge, ...)
  • Server clusters, networks of workstations via fast links
    (ATM, FDDI, ...)

• Problem persisted: tedious hand-coding needed: lightweight process / threads programming, low-level synchronization, etc.

• LP/CLP/CC: already offer high application development efficiency and good sequential performance (close to imperative languages).

• LP/CLP/CC, from the automatic parallelization point of view:
  • program close to problem description
  • well understood mathematical foundation
  • relative purity (side-effects, well behaved variable scoping)
  less hiding of intrinsic parallelism
  possibility of formal treatment
  more amenable to automatic parallelization.