Add Support for Predictive Scaling #49
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Also did a reformat (-> Palladio Code guidelines), that's why the change is so large
Aggregation now (reliably) supports Utilization + Operation Response Times. The Number of Elements stimulus does not work yet as the `MeasurementMade` event doesn't reach the Aggregator
- Use SlidingTimeWindowAggregation - Add RateOfChange aggregation
This performs a linear regression and thus needs to know the time at which datapoints have been measured - thus the Aggregations all have DataPoints as input now and the WindowAggregation classes have been changed accordingly
- Adds UtilizationRewardEvaluator - Adds new Functions to FunctionRewardEvaluator - Misc changes to interpreters
This caused the number of event handlers to grow exponentially and all sorts of problems
Currently, the average over the last interval is taken for non-aggregated stimuli
- Removed Reward entities - Added reward evaluation to LearningBasedModelEvaluator
For reasoning see metamodel
This just changes the behavior from default-false to default-true, more sophisticated behavior should probably be implemented
- Introduced convenience constructor - Better documentation - Support printing the trained model
- Improve readability by introducing new methods - Remove unneeded methods - Improve comments - Improve reward + add fuzzy membership parameters
- Introduces AbstractFuzzyLearningModelEvaluator as abstract parent class for Fuzzy SARSA + Q-Learning
- Add knowledge for initialization - Exploration bonus at beginning - Improved logging - Changed reward to discourage scaling out
Also changes the reward for Q-Learning, still WIP
For interval: IntervalConstraint is used For max container count: TargetGroupSizeConstraint is suggested
This should be better if the scaling action is somehow modified, e.g. by some filters enforcing some constraints such as a thrashing constraint
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR implements things needed for the model changes proposed in PalladioSimulator/Palladio-Addons-SPD-Metamodel#32, notably:
ModelInterpreterfor the model, which returns the correctModelEvaluator), modifies theSpdInterpreterto support the newModelBasedScalingPolicyand the changedReactiveScalingPolicyModelBasedTriggerCheckerwhich receivesMeasurementMadeandRepeatedSimulationTimeReachedevents and uses aModelEvaluatorto decide whether a scaling action should happen (a more detailed EPC diagram is given below)RepeatedSimulationTimeReachedevent is added that repeats in some given interval and is used byModels to perform periodic operations such as scaling decisions. This event is re-scheduled by a new listener called RepeatedSimulationTimeRepeaterRateOfChangeAggregationthat performs a regression (usingorg.apache.commons.math3.stat.regression) to find out the rate of change, requiring a change of the aggregator data structure to DataPoints that contain both the value and the time of the stimulus output that is being aggregatedThe
ModelEvaluatorsfor the different Models will be implemented in separate PRs (the current progress can be found in my fork) as this is already a quite large PR.Finally, here's an EPC depicting the interaction between
MeasurementMade/RepeatedSimulationTimeReachedevents andModelAdjustmentRequestedevents based on the constructs added for the newly introducedModelBasedScalingPolicy: