Changes between Version 19 and Version 20 of GpuWorkFetch


Ignore:
Timestamp:
Dec 29, 2008, 4:00:59 PM (16 years ago)
Author:
davea
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • GpuWorkFetch

    v19 v20  
    7979
    8080We'll continue to use the idea of '''long-term debt''' (LTD).
    81 LTD represents how much work is "owed" to each project.
     81LTD represents how much work (measured in device instance-seconds) is "owed" to each project.
    8282This increases over time in proportion to its resource share,
    8383and decreases as it uses resources.
     
    103103 * The "overall LTD", which is used in the work-fetch decision, is the sum of the resource LTDs, weighted by the speed of the resource (FLOPs per instance-second).
    104104
    105 
     105Next we need to specify exactly how LTD is maintained.
     106It's clear how it decreases; the question is, how is it increased?
     107We need to avoid situations where LTD increases without bound.
     108We propose the following:
     109
     110 * For each project P and resource R there is a boolean flag D(P, R) indication whether P should accumulate debt for R.  The idea is that if D(P,R) is true, then it's likely that P would supply a job for R if we asked it.
     111 * D(P, R) is initially false.
     112 * If P supplies a job for R, D(P,R) is set to true.
     113 * If we send P a request that doesn't return any jobs, then for each resource R for which req_seconds(R)>0, D(P,R) is set to false.
    106114
    107115=== Per-resource-type backoff ===