DataStallScenario
Data scenarios · Implementation: mlsysim/engine/scenario_definitions/data.py.
Construct this scenario with keyword inputs. The fields below show the types and defaults declared by this class.
A batch delivered in io_time against compute_time of compute, in sequence or overlapped.
Notes
Equation: Compose mlsysim.solvers.DataStallModel.
Assumptions: Compute and I/O durations describe one batch; throughput comparisons use ideal overlap and fixed I/O time.
Excludes: Queue transients, buffering limits, and partial overlap are absent.
References
Derivation: this scenario binds its input fields to the referenced implementation contracts: mlsysim.solvers.DataStallModel. The methods below identify any additional arithmetic.
Examples
>>> from mlsysim import Scenarios
>>> case = Scenarios.OverlappedInputStall
>>> type(case.result).__name__
'DataStallResult'Attributes
| Name | Description |
|---|---|
| compute_time | |
| io_time | |
| plateau | Maximum compute-only speedup once I/O becomes the bottleneck. Equation: |
| result | Evaluate this scenario through mlsysim.solvers.DataStallModel.solve. Equation: pass |
Methods
| Name | Description |
|---|---|
| relative_throughput | Throughput change after speeding only compute in a perfectly overlapped pipeline. |
relative_throughput
engine.scenario_types.DataStallScenario.relative_throughput(compute_speedup)Throughput change after speeding only compute in a perfectly overlapped pipeline. Equation: old max(compute_time, io_time) divided by the new maximum. Units: dimensionless; batch work and I/O time are fixed.
Overlapped batches per second when compute runs compute_speedup times faster, delivery fixed.
compute_time
Type: Quantity.
Declared default: required.
io_time
Type: Quantity.
Declared default: required.
result
engine.scenario_types.DataStallScenario.result
Evaluate this scenario through mlsysim.solvers.DataStallModel.solve. Equation: pass the stated fields to that implementation; its contract defines the calculation and feasibility tests. Units: each returned field follows the referenced result schema; no independent equation is evaluated by this wrapper.
plateau
engine.scenario_types.DataStallScenario.plateau
Maximum compute-only speedup once I/O becomes the bottleneck. Equation: max(compute_time, io_time) / io_time. Units: dimensionless; requires positive I/O time and ideal overlap.