Skip to main content

StragglerCostScenario

Reliability scenarios · Implementation: mlsysim/engine/scenario_definitions/reliability.py.

Construct this scenario with keyword inputs. The fields below show the types and defaults declared by this class.

One of n_devices in a synchronous job runs at slow_speed of its normal rate.

Notes

Equation: Use the dimensional products, ratios, or selection rules documented on the calculated properties below.

Assumptions: One slower worker stalls every synchronous step; worker counts, task work, and hourly prices remain fixed.

Excludes: Asynchronous execution and selective worker replacement are absent.

References

Derivation: the calculated properties below state their defining arithmetic and units. They are scenario accounting or input-selection rules, not independently measured relationships.

Examples

>>> from mlsysim import Scenarios
>>> case = Scenarios.HalfSpeedStraggler1K
>>> str(case.slow_step)
'2.0 s'

Attributes

Name Description
device_hour_rate
idle_cost One-hour fleet rental attributed to the lost throughput fraction. Equation: n_devices | | [n_devices](#mlsysim.engine.scenario_types.StragglerCostScenario.n_devices) | | | [slow_speed](#mlsysim.engine.scenario_types.StragglerCostScenario.slow_speed) | | | [slow_step](#mlsysim.engine.scenario_types.StragglerCostScenario.slow_step) | Synchronous step duration imposed by the slower worker. Equation:step_time /
step_time
throughput Throughput retained relative to the healthy step. Equation: step_time / slow_step.

n_devices

Type: int.

Declared default: required.

step_time

Type: Quantity.

Declared default: required.

slow_speed

Type: float.

Declared default: required.

device_hour_rate

Type: Quantity.

Declared default: required.

slow_step

engine.scenario_types.StragglerCostScenario.slow_step

Synchronous step duration imposed by the slower worker. Equation: step_time / slow_speed. Units: seconds; all workers wait for this one straggler.

throughput

engine.scenario_types.StragglerCostScenario.throughput

Throughput retained relative to the healthy step. Equation: step_time / slow_step. Units: dimensionless; work per step is unchanged.

idle_cost

engine.scenario_types.StragglerCostScenario.idle_cost

One-hour fleet rental attributed to the lost throughput fraction. Equation: n_devices * device_hour_rate * (1 - throughput) * one_hour. Units: USD; this is an allocated cost, not an extra bill.

Back to top