loadvariationriskbalancing

README

LoadVariationRiskBalancing Plugin

The LoadVariationRiskBalancing plugin is one of the Trimaran scheduler plugins and is described in detail in Trimaran: Real Load Aware Scheduling. The Trimaran plugins employ the load-watcher in order to collect measurements from the nodes as described here.

The (normalized) risk of a node is defined as a combined measure of the average and standard deviation of the node utilization. It is given by

risk = [ average + margin * stDev^{1/sensitivity} ] / 2

where average​ and stDev​ are the fractional (between 0 and 1) measured average utilization and standard deviation of the utilization over a period of time, respectively. The two parameters: margin​ and sensitivity​, impact the amount of risk due to load variation. In order to magnify the impact of low variations, the stDev​ quantity is raised to a fractional power with the sensitivity​ parameter being the root power. And, the margin​ parameter scales the variation quantity. The recommended values for the margin​ and sensitivity​ parameters are 1 and 2, respectively. Each of the two added terms is bounded between 0 and 1. Then, the divisor 2 is used to normalize risk between 0 and 1.

(Since the additional load due to the pod, that is the subject of scheduling, is not known in advance, we assume that its average and standard deviation load are the requested amount and zero, respectively.)

Risk is calculated independently for the CPU and memory resources on the node. Let worstRisk be the maximum of the two calculated risks. The score of the node, assuming that minScore is 0, is then computed as

score = maxScore * (1 - worstRisk)

Thus, the LoadVariationRiskBalancing plugin has the following configuration parameters:

• safeVarianceMargin : Multiplier (non-negative floating point) of standard deviation. (Default 1)
• safeVarianceSensitivity : Root power (non-negative floating point) of standard deviation. (Default 1)

In addition, we have the watcherAddress or metricProviderconfiguration parameters, depending on whether the load-watcher is in service or library mode, respectively.

Following is an example scheduler configuration with the LoadVariationRiskBalancing plugin enabled, and using the load-watcher in library mode, collecting measurements from the Prometheus server.

apiVersion: kubescheduler.config.k8s.io/v1beta2
kind: KubeSchedulerConfiguration
leaderElection:
  leaderElect: false
profiles:
- schedulerName: trimaran
  plugins:
    score:
      enabled:
       - name: LoadVariationRiskBalancing
  pluginConfig:
  - name: LoadVariationRiskBalancing
    args:
      safeVarianceMargin: 1
      safeVarianceSensitivity: 2
      metricProvider:
        type: Prometheus
        address: http://prometheus-k8s.monitoring.svc.cluster.local:9090

Documentation

Overview

Package loadvariationriskbalancing plugin attempts to balance the risk in load variation across the cluster. Risk is expressed as the sum of average and standard deviation of the measured load on a node. Typical load balancing involves only the average load. Here, we consider the variation in load as well, hence resulting in a safer balance.

Index

• Constants
• func New(obj runtime.Object, handle framework.Handle) (framework.Plugin, error)
• type Collector
• type LoadVariationRiskBalancing
  • func (pl *LoadVariationRiskBalancing) Name() string
  • func (pl *LoadVariationRiskBalancing) NormalizeScore(context.Context, *framework.CycleState, *v1.Pod, framework.NodeScoreList) *framework.Status
  • func (pl *LoadVariationRiskBalancing) Score(ctx context.Context, cycleState *framework.CycleState, pod *v1.Pod, ...) (int64, *framework.Status)
  • func (pl *LoadVariationRiskBalancing) ScoreExtensions() framework.ScoreExtensions

Constants

const (
	// Name : name of plugin
	Name = "LoadVariationRiskBalancing"
)

Functions

func New

func New(obj runtime.Object, handle framework.Handle) (framework.Plugin, error)

New : create an instance of a LoadVariationRiskBalancing plugin

Types

type Collector

type Collector struct {
	// contains filtered or unexported fields
}

Collector : get data from load watcher, encapsulating the load watcher and its operations

Currently, the Collector owns its own load watcher and is used solely by the LoadVariationRiskBalancing plugin. Other Trimaran plugins, such as the TargetLoadPacking, have their own load watchers. The reason being that the Trimaran plugins have different, potentially conflicting, objectives. Thus, it is recommended not to enable them concurrently. As such, they are currently designed to each have its own load-watcher. If a need arises in the future to enable multiple Trimaran plugins, a restructuring to have a single Collector, serving the multiple plugins, may be beneficial for performance reasons.

type LoadVariationRiskBalancing

type LoadVariationRiskBalancing struct {
	// contains filtered or unexported fields
}

LoadVariationRiskBalancing : scheduler plugin

func (*LoadVariationRiskBalancing) Name

func (pl *LoadVariationRiskBalancing) Name() string

Name : name of plugin

func (*LoadVariationRiskBalancing) NormalizeScore

func (pl *LoadVariationRiskBalancing) NormalizeScore(context.Context, *framework.CycleState, *v1.Pod, framework.NodeScoreList) *framework.Status

NormalizeScore : normalize scores

func (*LoadVariationRiskBalancing) Score

func (pl *LoadVariationRiskBalancing) Score(ctx context.Context, cycleState *framework.CycleState, pod *v1.Pod, nodeName string) (int64, *framework.Status)

Score : evaluate score for a node

func (*LoadVariationRiskBalancing) ScoreExtensions

func (pl *LoadVariationRiskBalancing) ScoreExtensions() framework.ScoreExtensions

ScoreExtensions : an interface for Score extended functionality