[poolifier.git] / src / pools / selection-strategies / README.md

# Worker choice strategies

All duration or timestamp are expressed in milliseconds.

## Strategies

### Fair share

Its goal is to distribute the load evenly across all workers. To achieve this, the strategy keeps track of the average task execution time for each worker and assigns the next task to the worker with the lowest task end prediction time: `task_end_prediction = max(current_time, task_end_prediction) + average_task_execution_time`.  
By default, the strategy uses the average task execution time for each worker but it can be configured to use the task event loop utilization (ELU) active time instead.

### Weighted round robin

The worker weights are maximum tasks execution time, once the worker has reached its maximum tasks execution time, the next task is assigned to the next worker. The worker default weights are the same for all workers and is computed given the CPU cores speed and theirs numbers.

### Interleaved weighted round robin

The worker weights are maximum tasks execution time. The rounds are the deduplicated worker weights.  
During a round, if worker weight is inferior to the current round, the next task is assigned to the next worker. Once all workers have been assigned a task, the next round starts.  
The worker default weights are the same for all workers and is computed given the CPU cores speed and theirs numbers. So the default rounds consists of a unique worker weight.

## Statistics

Worker choice strategies enable only the statistics that are needed to choose the next worker to avoid unnecessary overhead.

### Median

Strategies using the average task execution time for each worker can use the median instead. Median is more robust to outliers and can be used to avoid assigning tasks to workers that are currently overloaded. Median usage introduces a small overhead: measurement history must be kept for each worker and the median must be recomputed each time a task has finished.
Commit	Line	Data
9e2ab226 JB	1	# Worker choice strategies
9e2ab226 JB	2
c63b7b38 JB	3	All duration or timestamp are expressed in milliseconds.
c63b7b38 JB	4
9e2ab226 JB	5	## Strategies
	6
	7	### Fair share
	8
	9	Its goal is to distribute the load evenly across all workers. To achieve this, the strategy keeps track of the average task execution time for each worker and assigns the next task to the worker with the lowest task end prediction time: `task_end_prediction = max(current_time, task_end_prediction) + average_task_execution_time`.
f1c06930	10	By default, the strategy uses the average task execution time for each worker but it can be configured to use the task event loop utilization (ELU) active time instead.
9e2ab226 JB	11
	12	### Weighted round robin
	13
251f0d71	14	The worker weights are maximum tasks execution time, once the worker has reached its maximum tasks execution time, the next task is assigned to the next worker. The worker default weights are the same for all workers and is computed given the CPU cores speed and theirs numbers.
afe102b2	15
9e2ab226 JB	16	### Interleaved weighted round robin
9e2ab226 JB	17
251f0d71	18	The worker weights are maximum tasks execution time. The rounds are the deduplicated worker weights.
12d8ed1b	19	During a round, if worker weight is inferior to the current round, the next task is assigned to the next worker. Once all workers have been assigned a task, the next round starts.
251f0d71 JB	20	The worker default weights are the same for all workers and is computed given the CPU cores speed and theirs numbers. So the default rounds consists of a unique worker weight.
251f0d71 JB	21
9e2ab226 JB	22	## Statistics
9e2ab226 JB	23
afe102b2 JB	24	Worker choice strategies enable only the statistics that are needed to choose the next worker to avoid unnecessary overhead.
afe102b2 JB	25
9e2ab226 JB	26	### Median
9e2ab226 JB	27
c63b7b38	28	Strategies using the average task execution time for each worker can use the median instead. Median is more robust to outliers and can be used to avoid assigning tasks to workers that are currently overloaded. Median usage introduces a small overhead: measurement history must be kept for each worker and the median must be recomputed each time a task has finished.