MM1ServiceSpeedupScenario
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.
The same traffic on an M/M/1 server before and after its service time falls.
Notes
Equation: The calculation members below define arrival_rate, fast_utilization, slow_time, fast_time, slow_wait, fast_wait, service_gain, time_gain, wait_gain.
Assumptions: Arrival rate stays fixed while service time changes; stable M/M/1 queues require utilization below one.
Excludes: Burstiness, heavy-tailed service, and finite-queue rejection.
References
Derivation: The same traffic on an M/M/1 server before and after its service time falls. 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.ResNet50FasterModelAt80PctH100
>>> value = case.time_gain
>>> print(f"{value:.3g}")
5.87Attributes
| Name | Description |
|---|---|
| arrival_rate | Fixed offered request rate per second inferred from slow-service utilization. |
| fast_service | |
| fast_time | Fast-server mean time in system in ms at its reduced utilization. |
| fast_utilization | Dimensionless fast-server utilization at the same offered request rate. |
| fast_wait | Fast-server queueing wait in ms, mean time in system minus service time. |
| service_gain | Dimensionless slow-to-fast service-time ratio. |
| slow_service | |
| slow_time | Slow-server mean time in system in ms via calc_mm1_time_in_system. |
| slow_wait | Slow-server queueing wait in ms, mean time in system minus service time. |
| time_gain | Dimensionless slow-to-fast total-latency ratio, including queueing. |
| utilization | |
| wait_gain | Dimensionless slow-to-fast queueing-wait ratio. |
slow_service
Type: Quantity.
Declared default: required.
fast_service
Type: Quantity.
Declared default: required.
utilization
Type: float.
Declared default: required.
arrival_rate
engine.scenario_types.MM1ServiceSpeedupScenario.arrival_rate
Fixed offered request rate per second inferred from slow-service utilization.
Equation: (utilization / slow_service).to('1/second').
fast_utilization
engine.scenario_types.MM1ServiceSpeedupScenario.fast_utilization
Dimensionless fast-server utilization at the same offered request rate.
Equation: utilization * (fast_service / slow_service).to('').magnitude.
slow_time
engine.scenario_types.MM1ServiceSpeedupScenario.slow_time
Slow-server mean time in system in ms via calc_mm1_time_in_system.
Calculation: mlsysim.physics.serving.calc_mm1_time_in_system.
fast_time
engine.scenario_types.MM1ServiceSpeedupScenario.fast_time
Fast-server mean time in system in ms at its reduced utilization.
Calculation: mlsysim.physics.serving.calc_mm1_time_in_system.
slow_wait
engine.scenario_types.MM1ServiceSpeedupScenario.slow_wait
Slow-server queueing wait in ms, mean time in system minus service time.
Equation: (slow_time - slow_service).to('ms').
fast_wait
engine.scenario_types.MM1ServiceSpeedupScenario.fast_wait
Fast-server queueing wait in ms, mean time in system minus service time.
Equation: (fast_time - fast_service).to('ms').
service_gain
engine.scenario_types.MM1ServiceSpeedupScenario.service_gain
Dimensionless slow-to-fast service-time ratio.
Equation: (slow_service / fast_service).to('').magnitude.
time_gain
engine.scenario_types.MM1ServiceSpeedupScenario.time_gain
Dimensionless slow-to-fast total-latency ratio, including queueing.
Equation: (slow_time / fast_time).to('').magnitude.
wait_gain
engine.scenario_types.MM1ServiceSpeedupScenario.wait_gain
Dimensionless slow-to-fast queueing-wait ratio.
Equation: (slow_wait / fast_wait).to('').magnitude.