HBJ model

In computer science, the Helman-Bader-JaJa model [1] is a concise message-passing model of parallel computing defined with the following parameters:

  • p {\displaystyle p} is number of processors.
  • n {\displaystyle n} is the problem size.
  • m {\displaystyle m} is number of machine words in a packet sent over the network.
  • τ {\displaystyle \tau } is the latency, or time at which a processor takes to initiate a communication on a network.
  • σ {\displaystyle \sigma } is the bandwidth, or time per machine word at which a processor can inject or receive m {\displaystyle m} machine words from the network.
  • T c o m p {\displaystyle T_{comp}} is the largest computation time expended on a processor.
  • T c o m m {\displaystyle T_{comm}} is the time spent in communication on the network.

This model assumes that for any subset of q {\displaystyle q} processors, a block permutation among the q {\displaystyle q} processors takes ( τ + σ m ) {\displaystyle (\tau +\sigma m)} time, where m {\displaystyle m} is the size of the largest block.

Analysis of common parallel algorithms

Complexities of common parallel algorithms contained in the MPI libraries:[2]

  • Point to point communication: O ( τ + σ m ) {\displaystyle O(\tau +\sigma m)}
  • Reduction : O ( l o g ( p ) ( τ + σ m ) ) {\displaystyle O(log(p)(\tau +\sigma m))}
  • Broadcast: O ( l o g ( p ) ( τ + σ m ) ) {\displaystyle O(log(p)(\tau +\sigma m))}
  • Parallel prefix: O ( l o g ( p ) n p ( τ + σ m ) ) {\displaystyle O(log(p){n \over p}(\tau +\sigma m))}
  • All to all: O ( p ( τ + σ m ) ) ) {\displaystyle O(p(\tau +\sigma m)))}

References

  1. ^ David R., Helman; David A., Bader; JaJa, Joseph (1998). "A Randomized Parallel Sorting Algorithm with an Experimental Study" (PDF). Journal of Parallel and Distributed Computing. 52: 1–23. doi:10.1006/jpdc.1998.1462. hdl:1903/835. Retrieved 26 October 2012. [dead link]
  2. ^ Bader, David A.; Jaja, Joseph (1996). "Practical parallel algorithms for dynamic data redistribution, median finding, and selection". Proceedings of the 10th IEEE International Parallel Processing Symposium: 292–301.


  • v
  • t
  • e