ReplicaStartupScenario
Serving scenarios · Implementation: mlsysim/engine/scenario_definitions/serving.py.
Construct this scenario with keyword inputs. The fields below show the types and defaults declared by this class.
A serving replica’s startup phases; warm_path and cold_path name the phases each start pays.
Notes
Equation: The calculation members below define warm_total, cold_total.
Assumptions: Selected startup phases execute sequentially and each named phase is charged once per listed occurrence.
Excludes: Parallel phase overlap and contention-dependent startup durations.
References
Derivation: A serving replica’s startup phases; warm_path and cold_path name the phases each start pays. Each member records the corresponding algebra or named physics function.
Examples
Evaluate the named configuration and display a modeled result.
>>> from mlsysim import Scenarios
>>> case = Scenarios.ResNet50ReplicaStartup
>>> value = case.warm_total
>>> print(f"{value:.3g~}")
1.5 sAttributes
| Name | Description |
|---|---|
| cold_path | |
| cold_total | Sum of named cold-path phase durations in seconds, with sequential execution assumed. |
| phases | |
| ranges | |
| warm_path | |
| warm_total | Sum of named warm-path phase durations in seconds, with sequential execution assumed. |
phases
Type: Dict[str, Quantity].
Declared default: required.
ranges
Type: Dict[str, List[Quantity]].
Declared default: {}.
warm_path
Type: List[str].
Declared default: required.
cold_path
Type: List[str].
Declared default: required.
warm_total
engine.scenario_types.ReplicaStartupScenario.warm_total
Sum of named warm-path phase durations in seconds, with sequential execution assumed.
Equation: sum((phases[p] for p in warm_path), Q_(0, 'second')).to('second').
cold_total
engine.scenario_types.ReplicaStartupScenario.cold_total
Sum of named cold-path phase durations in seconds, with sequential execution assumed.
Equation: sum((phases[p] for p in cold_path), Q_(0, 'second')).to('second').