How a job used its calculator groups: the seconds each group spent inside engine calls, the systems it evaluated, and the driver's wall time inside grouped requests. More...
#include <CalculatorGroupUse.h>
Public Member Functions | |
| std::size_t | groups () const noexcept |
| double | meanBusy () const noexcept |
| double | maxBusy () const noexcept |
| double | loadBalance () const noexcept |
| Mean over max busy time: 1 when every group carried the same work. | |
| double | communicationEfficiency () const noexcept |
| Max busy over wall: the share of the driver's wait that the busiest group spent computing. | |
| double | parallelEfficiency () const noexcept |
| double | idleFraction (std::size_t group) const noexcept |
| Fraction of the grouped wall time a group spent outside engine calls. | |
| std::string | table () const |
Public Attributes | |
| std::vector< double > | busy |
| seconds in engine calls, per group | |
| std::vector< double > | systems |
| systems evaluated, per group | |
| double | wall {0.0} |
| driver seconds inside grouped requests | |
| double | singleWall {0.0} |
| of which single-system requests | |
| long | batches {0} |
| grouped batch requests | |
| long | singles {0} |
| single-system requests (group 0 only) | |
How a job used its calculator groups: the seconds each group spent inside engine calls, the systems it evaluated, and the driver's wall time inside grouped requests.
The ratios follow the POP parallel efficiency model: parallel efficiency is load balance times communication efficiency.
Definition at line 28 of file CalculatorGroupUse.h.
|
inlinenodiscardnoexcept |
Max busy over wall: the share of the driver's wait that the busiest group spent computing.
Waits for a slower group in one batch, serial requests on group 0 and the result broadcasts all lower it.
Definition at line 59 of file CalculatorGroupUse.h.
|
inlinenodiscardnoexcept |
Definition at line 36 of file CalculatorGroupUse.h.
|
inlinenodiscardnoexcept |
|
inlinenodiscardnoexcept |
Mean over max busy time: 1 when every group carried the same work.
Definition at line 51 of file CalculatorGroupUse.h.
|
inlinenodiscardnoexcept |
Definition at line 46 of file CalculatorGroupUse.h.
|
inlinenodiscardnoexcept |
|
inlinenodiscardnoexcept |
Definition at line 63 of file CalculatorGroupUse.h.
|
inlinenodiscard |
Definition at line 75 of file CalculatorGroupUse.h.
| long eonc::CalculatorGroupUse::batches {0} |
| std::vector<double> eonc::CalculatorGroupUse::busy |
seconds in engine calls, per group
Definition at line 29 of file CalculatorGroupUse.h.
| long eonc::CalculatorGroupUse::singles {0} |
| double eonc::CalculatorGroupUse::singleWall {0.0} |
| std::vector<double> eonc::CalculatorGroupUse::systems |
systems evaluated, per group
Definition at line 30 of file CalculatorGroupUse.h.
| double eonc::CalculatorGroupUse::wall {0.0} |