Changes between Version 3 and Version 4 of RuntimeEstimation


Ignore:
Timestamp:
May 11, 2012, 9:15:42 AM (13 years ago)
Author:
davea
Comment:

--

Legend:

Unmodified
Added
Removed
Modified

  • RuntimeEstimation

    v3 v4  
    88the scheduler estimates its FLOPS.
    99This is either the CPU benchmark,
    10 or a value assigned by the app_plan() function.
    11 
    12 The app_plan function is expected to predict
    13 the performance of an app on all possible hosts.
     10or a value assigned by the app_plan() function
     11(the app_plan function is expected to predict
     12the performance of an app on all possible hosts).
    1413
    1514The client maintains a per-project duration correction factor (DCF),
    16 which was intended to measure the efficiency of the project's apps,
     15intended to measure the efficiency of the project's apps,
    1716and the systematic error in wu.rsc_fpops_est.
    18 DCF was used to scale runtime estimates on both client and server side.
     17DCF is used to scale runtime estimates on both client and server side.
    1918
    2019Problems with the old system:
    2120
    2221 * Projects can have lots of apps.  A single DCF does not suffice.
    23  * Projects can't be expected to predict app performance.
     22 * Projects can't be expected to predict app performance,
     23   either in wu.rsc_fpops_est or in app_plan().
    2424
    2525== The new system ==