The Windowing field in a study specifies the windowing policy to be adopted to score the experiments of an optimization study.

The two available windowing strategies have different structures:

• Trim windowing: trim the temporal interval of a trial, both from the start and the end of a specified temporal amount - this is the default strategy

• Stability windowing: discard temporal intervals in which a given metric is not stable and selects the temporal interval in which a metric is maximized or minimized.

In case the windowing strategy is not specified, the entire time window is considered.

Optimization goals and constraints are defined using a YAML manifest with the following structure:

goal:
  objective: "minimize"
  function:
    formula: "jvm1.response_time + jvm2.response_time"
  constraints:
    relative:
      - name: memory_used_max
        formula: jvm1.memory_used <= 80%
    absolute:
      - name: heap_used_max
        formula: jvm1.heap_used <= 3221225472

where:

Function

The function field of the Goal of a Study details the characteristics of the function Akamas should minimize or maximize to reach the desired performance objective.

The function field has the following structure:

function:
  formula: "jvm1.response_time / sqrt(x)"
  variables:
    x:
      metric: "throughput"
      labels:
        componentName: "jvm2"
      aggregation: "MAX"

Where:

Formula

The formula field represents the mathematical expression of the performance objective for the Study and contains variables and operators with the following characteristics:

• Valid operators are: + - * / ^ sqrt(variable) log(variable) max(variable1, variable2) min(variable1, variable2)

• Valid variables are in the form:

  • <component_name>.<metric_name>, which correspond directly to metrics of Components of the System under test

  • <variable_name>, which should match variables specified in the variables field

Each metric that is directly or indirectly part of the formula of the function of the Goal is aggregated by default by average; more specifically, Akamas computes the average of each metric within the time window specified by the windowing strategy of the Study.

Variables

The variables field contains the specification of additional variables present in the formula, variables that can offer more flexibility with respect to directly specifying each metric of each Component in the formula.

Notice: each subfield of variables specifies a variable with its characteristics, the name of the subfield is the name of the variable.

The variable subfield has the following structure:

It is possible to use the notation <component_name>.<metric_name> in the metric field to automatically filter the metric’s data point by that component name is applied.

Constraints

The constraints field specifies constraints on the metrics of the components of the system under test that need to be satisfied for a configuration to be valid with respect to the defined goal.

Constraints always consider the average value of specified metrics within the time window specified by the windowing strategy of the Study.

Constrainst may be absolute or relative. Absolute constraints set a fixed limit that should not be overcome. Relative constraints set a percentage value (calculated from the baseline value after the baseline experiment runs). Then each constraint (absolute or relative) has a name and a formula. The formula has the same syntax as the goal formula so refer to "formula" section above.

Examples

The following example refers to a study whose goal is to optimize the throughput of a Java service (jpetstore), that is to maximize the throughput (measured as elements_per_second) while keeping errors (error_rate) and latency (avg_duration, max_duration) under control (absolute values):

goal:
    objective: "maximize",
    function:
      formula: "jpetstore.elements_per_second"
    constraints:
      absolute:
        - name: min_elements_per_sec
          formula: jpetstore.elements_per_second > 55    
        - name: max_duration
          formula: jpetstore.max_duration < 800    
        - name: min_elements_per_sec
          formula: jpetstore.elements_per_second > 55    
        - name: error_rate
          formula: jpetstore.error_rate < 0.01
      relative:
        - name: avg_duration
          formula: jpetstore.avg_duration < 5%

The following example refers to a study whose goal is to optimize the memory consumption of Docker containers in a microservices application, that is to minimize the average memory consumption of Docker containers within the application of appId="app1" by observing memory limits, also normalizing by the maximum duration of a benchmark (container_benchmark_duration).

goal:
  objective: "minimize"
  function:
    formula: "containers_memory_limit/containers_benchmark_duration"
    variables:
      containers_memory_limit:
        metric: "memory_limit"
        labels:
          appId: "app1"
      containers_job_duration:
        metric: "benchmark_duration"
        labels:
          appId: "app1"
        aggregation: "MAX"

The MetricSelection field in a study specifies the metrics of the system. Such selection has only the purpose to specify which metrics need to be tracked while running the study and does not affect the optimization.

In case this selection is not specified, all metrics are considered.

A metrics selection can either assume the value of all to indicate that all the available metrics of the system of the study should be tracked, or can assume the value of a list of the names of metrics of the system that should be tracked prepended with the name of the component.

Example

The following fragment is an example:

A windowing policy of type trim trims the temporal interval of a trial, both from the start and from the end of a specified temporal amount (e.g., 3 seconds).

The trim windowing has the following structure:

In case a windowing policy is not specified, the default windowing corresponding to trim[0s,0s] is considered.

Example

The following fragment shows a windowing strategy of type "trim" where the time window is specified to start 10s after the beginning of the trial and to end immediately before the end of the trial:

The ParameterSelection field in a study specifies which parameters of the system should be tuned while running the optimization study.

In case this selection is not specified, all parameters are considered.

A parameter selection can either assume the value of all to indicate that all the available parameters of the system of the study should be tuned, or can assume the value of a list with items of the shape like the one below:

Notice that, by default, every parameter specified in the parameters selection of a study is applied. This can be modified, by leveraging the options.

Example

The following fragment is an example:

A windowing policy of type stability discards temporal intervals in which a given metric is not stable, and selects, among the remaining intervals, the ones in which another target metric is maximized or minimized. Stability windowing can be sample-based or time-frame based.

The stability windowing has the following structure:

and for the comparison metric section

Example

The following fragment is an example of stability windowing (time-frame based):

The kpis field in a study specifies which metrics should be considered as KPI for an offline optimization study.

In case this selection is not specified, all metrics mentioned in the goal and constraint of the optimization study are considered.

A KPI is defined as follows:

Notice that the textual badge displayed in the Akamas UI use "Best name".

Example

The following fragment is an example of a list of KPIs:

kpis:
  - name: "Response time"
    formula: renaissance.response_time
    direction: minimize
  - name: "CPU used"
    formula: renaissance.cpu_used
    direction: minimize
  - name: "Memory used"
    formula: renaissance.mem_used
    direction: minimize

A baseline step performs an experiment (a baseline experiment) and marks it as the initial experiment of a study. The purpose of the step is to build a reference configuration that Akamas can use to measure the effectiveness of an optimization conducted towards a system.

When a bootstrap step imports an experiment from another study, the step copies not only the experiment but also its trials and the system metrics generated during its execution.

The bootstrap step has the following structure:

where the from field should have the following structure:

study: "study_bootstrap_1" 2experiments: [1,2,3]

with:

• study contains the name or id of the study from which to import experiments

• experiments contains the numbers of the experiments from which to import

Examples

The following is an example of a bootstrap step that imports four experiments from two studies:

name: "my_bootstrap" # name of the step
type: "bootstrap" # type of the step (bootstrap)
from:
  - study: "study_bootstap_1" # name or id of the study from which to import
    experiments: [1, 2, 4] # the numbers of the experiments to import
  - study: "study_bootstrap_2"
    experiments: [1]

You can also import all the experiments of a study by omitting the experiments field:

name: "my_bootstrap" # name of the step
type: "bootstrap" # type of the step (bootstrap)
from:
  - study: "study_bootstrap_1" # name or id of the study from which to import

The Steps field in a study specifies the sequence of steps executed while running the study. These steps are run in exactly the same order in which they will be executed.

The following types of steps are available:

• Baseline: performs an experiment and sets it as a baseline for all the other ones

• Bootstrap: imports experiments from other studies

• Preset: performs an experiment with a specific configuration

• Optimize: performs experiments and generates optimized configurations

Notice that the structure corresponding to the steps is different for the different types of steps.

Optimization studies are defined using a YAML manifest with the following structure:

system: 1
name: Optimizing the e-shop application
goal:
  objective: maximize
  function:
    formula: payments_per_sec
    variables:
      payments_per_sec:
        metric: eshop_payments
        labels:
          componentName: eshop

workflow: eshop_jmeter_test
steps:
  - name: baseline
    type: baseline
    values:
      tomcat.maxThreads: 1024
      jvm.maxHeap: 2048
      jvm.garbageCollectorType: G1GC
      postgres.shared_buffers: 4096

with the following mandatory properties:

Some of these optional properties depend on whether the study is an offline or live optimization study.

Number of trials

It is possible to perform more than one trial per experiment in order to validate the score of a configuration under test, e.g. to take into account noisy environments.

The following fragment of the YAML definition of a study sets the number of trials to 3:

numberOfTrials: 3

Notice: This is a global property of the study which can be overwritten for each step.

Trial aggregation

The trial aggregation policy defines how trial scores are aggregated to form experiment scores.

There are three different types of strategies to aggregate trial scores:

• AVG: the score of an experiment is the average of the scores of its trials - this is the default

• MIN: the score of an experiment is the minimum among the scores of its trials

• MAX: the score of an experiment is the maximum among the scores of its trial

The following fragment of the YAML definition of a study sets the trial aggregation to MAX:

trialAggregation: MAX # other possible values are AVG, MIN

Examples

The following system refers to an offline optimization study for a system modeling an e-commerce service, where a windowing strategy is specified:

system: "bde4f259-9a51-4c67-87aa-3c5bc599c6b9" # id of the system to optimize with the actions defined in this study
workflow: "eshop_jmeter_test" # name of the workflow to use to perform trials
name: Optimizing the e-shop application # name of the study
goal: # the performance goal to achieve
  objective: "maximize"
  function:
    formula: "eshop.payments_per_second"
windowing: # the temporal window in which to compute the score of a trial
  type: "trim"
  trim: ["10s", "0s"] # use the duration of the trial minus 0s from start and end to compute the score of the trial
parametersSelection: "all" # use all available configuration parameters
metricsSelection: "all" # gather all metrics
steps: # the steps to conduct to perform experiments and trials
  - name: "my_baseline" # do first a baseline with the provided configuration
    type: "baseline"
    values:
      jvm.maxHeap: 2048
      jvm.gcType: "-XX:+UseParallelGC"
  - name: my_optimization # then do 20 optimization experiments of 2 trials each
    type: optimize
    numberOfExperiments: 200
    numberOfTrials: 2

The following offline study refers to a tuning initiative for a Cassandra-based system (ID 2)

system: 2
name: Optimizing the cassandra - team 2
goal:
  objective: minimize
  function:
    formula: read_response_time_p90
    variables:
      read_response_time_p90:
        metric: read_response_time_p90
        labels:
          componentName: cassandra

windowing:
  type: trim
  trim: [5m, 1m]

workflow: cassandra_workflow
parametersSelection:
  - name: cassandra_jvm.jvm_maxHeapSize
  - name: cassandra.cassandra_concurrentReads
  - name: cassandra.cassandra_concurrentWrites
  - name: cassandra.cassandra_fileCacheSizeInMb
  - name: cassandra.cassandra_memtableCleanupThreshold
  - name: cassandra.cassandra_concurrentCompactors

steps:
  - name: baseline_step
    type: baseline
    values:
      cassandra_jvm.jvm_maxHeapSize: 1024
      cassandra.cassandra_concurrentReads: 32
      cassandra.cassandra_concurrentWrites: 32
      cassandra.cassandra_fileCacheSizeInMb: 512
      cassandra.cassandra_memtableCleanupThreshold: 0.11
      cassandra.cassandra_concurrentCompactors: 2

  - name: optimization_step
    type: optimize
    optimizer: CALABI
    numberOfExperiments: 50

The following offline study is for tuning another Cassandra-based system (ID 3) by acting only on JVM and Linux parameters

system: 3
name: Optimizing a Cassandra NoSQL database version 3 (jvm + os parameters)
goal:
  objective: minimize
  function:
    formula: (x1+x2)/2
    variables:
      x1:
        metric: write_response_time_p90
        labels:
          componentName: cassandra_team1
      x2:
        metric: read_response_time_p90
        labels:
          componentName: cassandra_team1

windowing:
  type: trim
  trim: [8m,2m]

numberOfTrials: 2
workflow: cassandra_workflow_jvm_os

parametersSelection:
  - name: JVM1.jvm_maxHeapSize
  - name: JVM1.jvm_newRatio
  - name: JVM1.jvm_survivorRatio
  - name: JVM1.jvm_maxTenuringThreshold
  - name: JVM1.jvm_gcType
  - name: JVM1.jvm_concurrentGCThreads
  - name: os1.os_cpuSchedMinGranularity
  - name: os1.os_cpuSchedWakeupGranularity
  - name: os1.os_CPUSchedMigrationCost
  - name: os1.os_CPUSchedChildRunsFirst
  - name: os1.os_CPUSchedLatency

steps:
  - name: baseline_step
    type: baseline
    values:
      JVM_team1.jvm_maxHeapSize: 1024
      JVM_team1.jvm_newRatio: 2
      JVM_team1.jvm_survivorRatio: 8
      JVM_team1.jvm_maxTenuringThreshold: 15
      JVM_team1.jvm_gcType: UseConcMarkSweepGC
      JVM_team1.jvm_concurrentGCThreads: 8
      os_team1.os_cpuSchedMinGranularity: 3000000
      os_team1.os_cpuSchedWakeupGranularity: 4000000
      os_team1.os_CPUSchedMigrationCost: 500000
      os_team1.os_CPUSchedChildRunsFirst: 0
      os_team1.os_CPUSchedLatency: 24000000

  - name: optimization_sobol
    type: optimize
    optimizer: SOBOL
    numberOfExperiments: 3

  - name: optimization_calabi
    type: optimize
    optimizer: CALABI
    numberOfExperiments: 50

A preset step performs a single experiment with a specific configuration. The purpose of this step is to help you quickly understand how good is a particular configuration.

A preset step offers two options when selecting the configuration of the experiment to be executed:

• Use a configuration taken from an experiment of a study (can be the same study)

• Use a custom configuration

The preset step has the following structure:

where the from field should have the following structure:

study: "study_preset_1"
experiments: [1]

with

• study contains the name or id of the study from which to take the configuration. in case this is omitted, the same study of the step is considered for experiments from which taking configurations

• experiments contains the number of the experiment from which to take the configuration

Examples

Custom configuration

You can provide a custom configuration by setting values:

name: "my_preset" # name of the step
type: "preset" # type of the step (preset)
values:
  jvm1.maxHeapSize: 1024 # parameter maxHeapSize of jvm1 is set to 1024
  jvm2.maxHeapSize: 2048 # parameter maxHeapSize of jvm2 is set to 2048

Configuration from another study

You can select a configuration taken from another study by setting from:

name: "my_preset"  # name of the step
type: "preset" # type of the step (preset)
from:
  - study: "preset_study_!" # name or id of the study from which to take the configuration
    experiments: [1] # the number of the experiment from which to take the configuration

Configuration from the same study

You can select a configuration taken from the same study by setting from but by omitting the study field:

name: "my_preset"  # name of the step
type: "preset" # type of the step (preset)
from:
  - experiments: [1] # the step will take the configuration of the experiment #1 of the same study of the step

Notice: the from and experiments fields are defined as an array, but can only contain one element.

The renderParameters and doNotRenderParameters can be used to specify which configuration parameters should be rendered when doing experiments within a step.

Parameter rendering can be defined at the step level for baseline, preset, and optimize steps. This is not possible for bootstrap steps as bootstrapped experiments are not executed.

Examples

The following baseline step specifies that every parameter of the component 'os' should not be rendered while the parameter 'cpu_limit' of the component 'docker' should be rendered:

The following preset step specifies that the parameter 'cpu_limit' of the component 'docker' should be rendered:

The following optimize step specifies that every parameter of the component 'os' should not be rendered:

The workloadsSelection is a structure used to define the metrics that are used by Akamas to model workloads as part of a live optimization study.

with the following fields:

Notice that, while the domain can be used to specify the maximum and minimum values for a workload metric, it is not recommended to specify static workloads domains as Akamas can dynamically define them by using the observed minimum and maximum value of the metric.

Also notice that workload metrics must have been defined in the metricsSelection.

Examples

The following refers to a workload represented by the metric transactions_throughput of the konakart component:

A baseline step performs an experiment (a baseline experiment) and marks it as the initial experiment of a study. The purpose of the step is to build a reference configuration that Akamas can use to measure the effectiveness of an optimization conducted towards a system.

A baseline step offers three options when it comes to selecting the configuration of the baseline experiment:

• Use a configuration made with the default values of the parameters taken from the system of the study

• Use a configuration taken from an experiment of another study

• Use a custom configuration

The baseline step has the following structure:

where the from field should have the following structure:

with

• study contains the name or id of the study from which to take the configuration

• experiments contains the number of the experiment from which to take the configuration

Examples

Baseline configuration with default values

Default values for the baseline configuration only require setting the name and type fields:

Baseline configuration from another study

The configuration taken from another study to be used as a baseline only requires setting the from field:

Notice: the from and experiments fields are defined as an array, but can only contain one element.

Custom baseline configuration

The custom configuration for the baseline only requires setting the values field:

An optimize step generates optimized configurations according to the defined optimization strategy. During this step, Akamas AI is used to generate such optimized configurations.

The optimize step has the following structure:

Optimizer

The optimizer field allows selecting the desired optimizer:

• AKAMAS identifies the standard AI optimizer used by Akamas

• SOBOL identifies an optimizer that generates configurations using

• RANDOM identifies an optimization that generates configurations using random numbers

Notice that SOBOL and RANDOM optimizers do not perform initialization experiments, hence the field numberOfInitExperiments is ignored.

Optimizer options for offline studies

For offline optimization studies, the optimizerOptions field can be used to specify whether beta-warping optimization (a more sophisticated optimization that requires longer time) should be used for how many experiments (as a percentage):

where experimentsWithBeta can be:

• A percentage between 0 and 100%

• A number less than or equal to numberOfExperiments

Optimizer options for live studies

For live optimization studies, the optimizerOptions field can be used to specify several important parameters governing the live optimization, which can be defined at the study level and also overridden at the step level (only for steps of type optimize):

Online Mode

The onlineMode field specifies how the Akamas optimizer should operate:

• RECOMMEND: configurations are recommended to the user by Akamas and are only applied after having been approved (and possibly modified) by the user;

• FULLY AUTONOMOUS MODE: configurations are immediately applied by Akamas.

Safety Mode

The safetyMode field describes how the Akamas optimizer should evaluate the goal constraints on a candidate configuration for that configuration to be considered valid:

• GLOBAL: the constraints must be satisfied by the configuration under all observed workloads in the configuration history - this is the value taken in case onlineMode is set to RECOMMEND;

• LOCAL: the constraints are evaluated only under the workload selected according to the workload strategy - this should be used with onlineMode set to FULLY_AUTONOMOUS.

Notice that when setting the safetyMode to LOCAL, the recommended configuration is only expected to be good for the specific workload selected under the defined workload strategy, but it might violate constraints under another workload.

Workload Strategy

The workloadOptimizedForStrategy field specifies the workload strategy that drives how Akamas leverages the workload information when looking for the next configuration:

• MAXIMIN: the optimizer looks for a configuration that maximizes the minimum improvements for all the already observed workloads;

• MEDIAN: for each workload, a median of all its values is considered - this works well to find a configuration that is good for the median of all the workloads;

• MOST_VIOLATED: for each workload, the workload of the configuration which results in most violations is considered.

Safety Factor

The safetyFactor field specifies how much the optimizer should stay on the safe side in evaluating a candidate configuration with respect to the goal constraints. A higher safety factor corresponds to a safer configuration, that is a configuration that is less likely to violate goal constraints.

Acceptable values are all the real values ranging between 0 and 1, with (safetyFactor - 0.5) representing the allowed margin for staying within the defined constraint:

• 0 means "no safety", as with this value the optimizer totally ignores goal constraints violations;

• 0.5 means "safe, but no margin", as with this value the optimizer only tries configurations that do not violate the goal constraints, by remaining as close as possible to them;

• 1 means "super safe", as with this value the optimize only tries configurations that are very far from goal constraints.

For live optimization studies, 0.6 is the default value, while for offline optimization studies, the default value is 0.5.

Exploration Factor

The explorationFactor field specifies how much the optimizer explores the (unknown) optimization space when looking for new configurations. For any parameter, this factor measures the delta between already tried values and the value of a new possible configuration. A higher exploration factor corresponds to a broader exploration of never tried before parameter values.

Acceptable values are all the real values ranging between 0 and 1, plus the special string FULL_EXPLORATION:

• 0 means "no exploration", as with this value the optimizer chooses a value among the previously seen values for each parameter;

• 1 means "full exploration, except for categories", as with this value the optimizer for a non-categorical parameter any value among all its domain values can be chosen, while only values (categories) that have already been seen in previous configurations are chosen for a categorical parameter;

• FULL_EXPLORATION means "full exploration, including categories" as with this value the optimizer chooses any value among all its domain values, including categories, even if not already seen in previous configurations.

Failures

An optimize step is fault-tolerant and tries to relaunch experiments when they fail. Nevertheless, the step imposes a limit on the number of failed experiments: if too many experiments fail, then the entire step fails too. By default, at most 30 experiments can fail while Akamas is optimizing systems. An experiment is considered failed when it either failed to run (i.e., there is an error in the workflow) or violated some constraint.

Inizialitations

An optimize step launches some initialization experiments (by default 10) that do not apply the AI optimizer and are used to find good configurations. By default, the step performs 10 initialization experiments.

Initialization experiments take into account bootstrapped experiments, experiments executed in preset steps, and baseline experiments.

Examples

The following fragment refers to an optimization study that runs 50 experiments using the SOBOL optimizer:

The following fragment refers to an optimization where 50% of the experiments need to use the beta-warping option, which enables a more sophisticated but longer optimization: