Changes between Version 23 and Version 24 of GpuWorkFetch


Ignore:
Timestamp:
Jan 25, 2009, 8:32:27 PM (16 years ago)
Author:
davea
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • GpuWorkFetch

    v23 v24  
    77 * If there's a CPU shortfall, request work from the project with highest long-term debt (LTD).
    88
    9 The scheduler request has a scalar "work_req_seconds"
     9The scheduler request has a single "work_req_seconds"
    1010indicating the total duration of jobs being requested.
    1111
    12 This policy has some problems.  First:
     12This policy has some problems:
    1313
    1414 * There's no way for the client to say "I have N idle CPUs; send me enough jobs to use them all".
     
    3838== Terminology ==
    3939
    40 New abstraction: '''processing resource''' or PRSC.
    41 The CPU is a PRSC and each coprocessor type is a PRSC.
     40New abstraction: '''processing resource type''' or PRT.
     41CPU and each coprocessor type are PRTs.
    4242
    4343A job sent to a client is associated with an app version,
     
    5454New fields in the scheduler request message:
    5555
    56  '''double cpu_req_seconds''':: number of CPU seconds requested
    57  '''double cpu_req_ninstances''':: send enough jobs to occupy this many CPUs
     56 '''double cpu_req_secs''':: number of CPU seconds requested
     57 '''double cpu_req_instances''':: send enough jobs to occupy this many CPUs
    5858
    5959And for each coprocessor type:
    6060
    61  '''double req_seconds''':: number of instance-seconds requested
    62  '''double req_ninstances''':: send enough jobs to occupy this many instances
     61 '''double req_secs''':: number of instance-seconds requested
     62 '''double req_instances''':: send enough jobs to occupy this many instances
    6363
    6464For compatibility with old servers, the message still has '''work_req_seconds''',