Changes between Version 5 and Version 6 of ValidationSummary


Ignore:
Timestamp:
Apr 14, 2010, 8:39:45 PM (14 years ago)
Author:
davea
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • ValidationSummary

    v5 v6  
    1 = Validation, credit, and replication =
     1= Validation and replication =
    22
    3 The execution of a job produces:
    4 
    5  * The output files;
    6  * The amount of CPU time used; this may be used to determine how much credit to grant for the result.
    7 
    8 In general, neither of these can be trusted, because:
     3The results of a job cannot be trusted, because:
    94
    105 * Some hosts have consistent or sporadic hardware problems, typically causing errors in floating-point computation.
    116 * Some volunteers may maliciously return wrong results; they may even reverse-engineer your application, deciphering and defeating any internal validation mechanism it might contain.
    12  * Some volunteers may return correct results but falsify the CPU time.
    137
    14 BOINC offers several mechanisms for validating results and credit.
     8BOINC offers several mechanisms for validating results.
    159However, there is no "one size fits all" solution.
    1610The choice depends on your requirements, and on the nature of your applications
     
    3428of detecting wrong results with high probability.
    3529
    36 The credit question remains.
    37 Some possibilities:
    38 
    39  * Grant fixed credit (feasible if your jobs are uniform).
    40  * Put a cap on granted credit (this allows cheating).
    41  * If claimed credit exceeds a threshold, replicate the job.
    42 
    4330== Replication ==
    4431
     
    4633each job gets done on N different hosts,
    4734and a result is considered valid if a strict majority of hosts return it.
    48 
    49 Replication also provides a good solution to credit cheating, even for non-uniform apps:
    50 grant the average claimed credit, throwing out the low and high first
    51 (if N=2, grant the minimum).
    5235
    5336One problem with replication is that there are discrepancies