34 | | approximately constant. |

35 | | |

36 | | The way to experiment is to use the setup as is, and change the |

37 | | lrefine_max in flash.par to get to the next problem size. Keep adjusting the number of |

38 | | processors until you get roughly the same number of blocks per processor. You may have less of a problem |

39 | | doing this on XT4 since there is lots more memory per processor. Also, eliminate IO |

| 34 | approximately constant. In simpler problems, it is possible to just change the |

| 35 | lrefine_max in flash.par to get to the next problem size. Then keep adjusting the number of |

| 36 | processors until you get roughly the same number of blocks per processor. |

| 37 | However, with WD_Def the amount of refinement is hard to calculate when varying |

| 38 | lrefine_max. Therefore, we vary r_match & uni_radius in the flash.pars |

| 39 | to adjust the problem size. We do not vary the refinement maximum level. |

| 40 | It required quite a bit of work to come up with these values. |

| 41 | They were chosen to give problem sizes of (roughly) 100 * 64 * 2^n |

| 42 | total blocks, for integer n, n is the number in scaling<n>. |

| 43 | |

| 44 | Note, weak scaling should be easier on the |

| 45 | XT4 since there is lots more memory per processor. Also, eliminate IO |

41 | | |

42 | | ~~However, it seems for WD_def we do not vary lrefine_max. ~~ |

43 | | ~~Chris needs to speak to his work colleague to understand this ~~ |

44 | | ~~more thoroughly. For the time being here is the email ~~ |

45 | | ~~describing the weak scaling for WD_Def:~~ |

46 | | ~~{{{~~ |

47 | | ~~On Wed, 1 Oct 2008, Chris Daley wrote:~~ |

48 | | |

49 | | ~~> Also can you please explain why we change:~~ |

50 | | ~~> r_match~~ |

51 | | ~~> refine_uni_radius~~ |

52 | | ~~> in each flash.par?~~ |

53 | | |

54 | | ~~THAT is how the size of the problem (i.e., muber of blocks) was~~ |

55 | | ~~changed. Note the lrefine_min and lrefine_max do not change!~~ |

56 | | |

57 | | ~~> How do you know the correct value?~~ |

58 | | |

59 | | ~~It required quite a bit of work to come up with these values.~~ |

60 | | |

61 | | ~~They were chosen to give problem sizes of (roughly) 100 * 64 * 2^n~~ |

62 | | ~~total blocks, for integer n, n is the number in scaling<n> I believe.~~ |

63 | | |

64 | | ~~Klaus ~~ |

65 | | ~~}}}~~ |

66 | | |