Changes between Version 8 and Version 9 of GpuWorkFetch


Ignore:
Timestamp:
Dec 26, 2008, 11:26:46 AM (16 years ago)
Author:
davea
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • GpuWorkFetch

    v8 v9  
    138138'''backoff timer'''*:  how long to wait until ask project for work only for this rsc
    139139double this any time we ask only for work for this rsc and get none
    140 (maximum 24 hours)
    141 clear it when we have a job that uses the rsc
     140(maximum 24 hours).
     141Clear it when we have a job that uses the PRSC.
    142142
    143143'''double share''': # of instances this project should get based on RS
     
    145145'''double long_term_debt*'''
    146146
    147 
     147=== Per-resource-type backoff
     148
     149We need to handle the situation where there's a GPU shortfall
     150but no projects are supplying GPU work
     151(for either permanent or transient reasons).
     152We don't want an overall backoff from those projects.
     153Insteac, we maintain separate backoff timer per PRSC.
    148154
    149155=== debt accounting ===
     
    158164do simulation as current
    159165on completion of an interval dt
    160         cpu_work_fetch.accumulate_shortfall(dt)
    161         cuda_work_fetch.accumulate_shortfall(dt)
    162 }}}
    163 
    164 --------------------
    165 scheduler request msg
    166 double work_req_seconds
    167 double cuda_req_seconds
    168 bool send_only_cpu
    169 bool send_only_cuda
    170 double ninstances_cpu
    171 double ninstances_cuda
    172 
    173 --------------------
    174 work fetch
    175 
    176 We need to deal w/ situation where there's GPU shortfall
    177         but no projects are supplying GPU work.
    178         We don't want an overall backoff from those projects.
    179         Solution: maintain separate backoff timer per resource
    180 
     166   cpu_work_fetch.accumulate_shortfall(dt)
     167   cuda_work_fetch.accumulate_shortfall(dt)
     168}}}
     169
     170
     171=== Work fetch ===
     172
     173{{{
    181174send_req(p)
    182175        switch cpu_work_fetch.priority
     
    203196                        send_req(p)
    204197                        return
    205 
    206 --------------------
    207 When get scheduler reply
    208         if request.
    209 --------------------
     198}}}
     199
     200=== Handling scheduler reply ===
     201
     202if request.
     203
    210204scheduler
    211205}}}