Changes between Version 12 and Version 13 of AlternateDistributionAlgorithms


Ignore:
Timestamp:
07/10/13 10:54:44 (12 years ago)
Author:
trac
Comment:

—

Legend:

Unmodified
Added
Removed
Modified

  • AlternateDistributionAlgorithms

    v12 v13  
    33This page is summarizes the communication and distribution methods used by other codes.
    44
    5  * [[latex(\(l\))]] is a specific level.
    6  * [[latex(\(g(l)\))]] is the number of grids on level [[latex(\(l\))]].
    7  * [[latex(\(p\))]] is the number of processors.
     5 * [[latex($\(l\)$)]] is a specific level.
     6 * [[latex($\(g(l)\)$)]] is the number of grids on level [[latex($\(l\)$)]].
     7 * [[latex($\(p\)$)]] is the number of processors.
    88
    99=== CASTRO ===
    … …  
    1111Although CASTRO can run both on pure MPI and an OpenMP-MPI hybrid, this discussion will concern itself with pure MPI.
    1212
    13  * At least one grid on each level is distributed to each core, meaning that [[latex(\(\frac{g(l)}{l} >= 1\))]].
     13 * At least one grid on each level is distributed to each core, meaning that [[latex($\(\frac{g(l)}{l} >= 1\)$)]].
    1414
    1515 * Each level has an associated !MultiFab object, holding the data from all the grids on that level.
    … …  
    2020  * Crutchfield's Knapsack algorithm, (see Crutchfield '91, Rendleman '00).
    2121  * Morton-ordering space-filling curve.
    22   * Dynamic switching between methods, based on [[latex(\(g\))]] and [[latex(\(p\))]].
     22  * Dynamic switching between methods, based on [[latex($\(g\)$)]] and [[latex($\(p\)$)]].
    2323
    2424[[BR]]
    … …  
    2727 * Grid hierarchies are replicated on each processor.
    2828
    29  * Root grid is split into [[latex(\(N_{patches} >= p\))]] and refined grids remain on their processors until the structure of the level is complete.
     29 * Root grid is split into [[latex($\(N_{patches} >= p\)$)]] and refined grids remain on their processors until the structure of the level is complete.
    3030
    3131 * Acknowledges that their distribution algorithm is less than optimal (references to a grid-splitting technique by Lan, Taylor & Bryan, '01.
    3232
    33  * Sibling search is sped up by using a coarse "chaining" mesh to map grids to neighbor lists.  Reduces the number of neighbor comparisons from [[latex(\(O(g^{2})\))]] to [[latex(\(O(g)\))]].
     33 * Sibling search is sped up by using a coarse "chaining" mesh to map grids to neighbor lists.  Reduces the number of neighbor comparisons from [[latex($\(O(g^{2})\)$)]] to [[latex($\(O(g)\)$)]].
    3434
    3535[[BR]]