Architectures for parallel computing

CS 300 (PDC)

Architectures for parallelism

• Three main resource categories for parallel computing:

  • processors (processing elements)

  • memories (storage)

  • communication (network, IC)

• Some example architectures

  • distributed computing system: multiple computers communicate via network to work together on applications

  • client-server: distributed system in which some computers function as clients and some as server(s) for particular applications

  • cluster: multiple computers on single network under single admin structure; typically dedicated, similar or same platform

  • grid: multiple computers on multiple networks, allocated by some authority for computing tasks; typically multiple platforms, multiple admin structures.

  • cloud: remote high-capacity computing resources; typically located at an unknown data centers, single admin structure, consisting of clusters

  • processor-level

  • multiprocessor: multiple PEs in a single "computer"

  • ______

  • ______

  • ______

  ____________
  ______

• multicore, manycore, hypercube, array, vector, ...

• Computational resources: processing elements, local memories, caches, global memories, IC network**, software licenses,

• Remote vs. local computations

• ______

• ______

• ______

• ______

• ______

rab@stolaf.edu, February 19, 2013