wrench::ParallelModel

class wrench::ParallelModel

A virtual class (with convenient static methods) to define parallel task performance models.

Subclassed by wrench::AmdahlParallelModel, wrench::ConstantEfficiencyParallelModel, wrench::CustomParallelModel

Public Functions

virtual double getParallelPerThreadWork(double total_work, unsigned long num_threads) = 0

A method that computes the amount of per-thread parallel work.

Parameters
  • total_work – the total amount of work (in flops)

  • num_threads – the number of threads

Returns

an amount of work (in flop)

virtual double getPurelySequentialWork(double total_work, unsigned long num_threads) = 0

A method that computes the amount of purely sequential work.

Parameters
  • total_work – the total amount of work (in flops)

  • num_threads – the number of threads

Returns

an amount of work (in flop)

Public Static Functions

static std::shared_ptr<ParallelModel> AMDAHL(double alpha)

Create an instance of an “Amdahl” parallel model.

Parameters

alpha – the fraction (i.e., a number between 0.0 and 1.0) of the task’s work that is perfectly parallelizable. Setting this value to 0 means that the task is purely sequential, and setting it to 1 means that the task is perfectly parallelizable.

Returns

a model instance

static std::shared_ptr<ParallelModel> CONSTANTEFFICIENCY(double efficiency)

Create an instance of a “Constant Efficiency” parallel model.

Parameters

efficiency – the parallel efficiency (which does not depend on the number of threads/cores).

Returns

a model instance

static std::shared_ptr<ParallelModel> CUSTOM(std::function<double(double, long)> lambda_sequential, std::function<double(double, long)> lambda_per_thread)

Create an instance of a “Custom” parallel model.

Parameters
  • lambda_sequential – a function that, when given a total flop amount and a number of threads, returns the amount of purely sequential work, in flops

  • lambda_per_thread – a function that, when given a total flop amount and a number of threads, returns the amount of per-thread parallel work, in flops

Returns

a model instance