welcome to seng 480b / csc 485a / csc 586a self-adaptive...

Welcome toSENG 480B / CSC 485A / CSC 586A

Self-Adaptive and Self-Managing Systems

Dr. Hausi A. MüllerDepartment of Computer Science

University of Victoria

http://courses.seng.uvic.ca/courses/2015/summer/seng/480ahttp://courses.seng.uvic.ca/courses/2015/summer/csc/485ahttp://courses.seng.uvic.ca/courses/2015/summer/csc/586a

1

Announcements Midterm II Thu, July 16 in class

A3 Posted and due July 10 Groups posted albeit incomplete

Grad project Handed out June 24 Due July 25

2

Autonomic Computing Vision

3

Autonomic Computing is really about making systemsself-managing …

—Paul Horn, IBM Research, 2001

New Reading Assignment Kephart & Walsh; An AI Perspective on AC Policies, 5th

IEEE International Workshop on Policies for Distributed Systems and Networks (Policy 2004)https://www.researchgate.net/publication/220726111_An_Artificial_Intelligence_Perspective_on_Autonomic_Computing_Policieshttp://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=1309145&punumber%3D9150%26filter%3DAND%28p_IS_Number%3A29053%29

John Wilkes, HP Labs: Utility functions, prices, and negotiation, HP Tech Report and Slides (2008)http://www.e-wilkes.com/john/papers/HPL-2008-81.pdf

4

MAPE-K LoopStandards & Interfaces

5

Events

Symptoms

Policies

Scripts

Script Interpreter

Symptom Database

Policy Types forAutonomic Computing Action policies

If-then action rules specify exactly what to do under the current condition.

Rational behaviour is compiled in by the designer Basis for reflex agents

Goal policies Requires self-model, planning,

conceptual knowledge representation Utility function policies

It chooses the actions to maximize its utility function Finer distinction between desriability of different states than goals Numerical characterization of state Needs methods to carry out actions to optimize utility 6

Utility Theory with SLAsJohn Wilkes, HP Labs: Utility functions, prices, and negotiation, HP Tech Report and Slides (2008)http://www.e-wilkes.com/john/papers/HPL-2008-81.pdfAbstract This paper provides an introduction to the use of utility theory with

service level agreements between a computer-based service provider and a client.

It argues that a consistent approach to utility, together with a flexible notion of pricing, can go a long way to clarifying some of the hidden assumptions that pervade many existing contracts and decisions around them.

The goal is to enhance understanding of a surprisingly tricky area, identify a few consequences for services providers and their clients, suggest a set of terminology that reduces ambiguities, and make some suggestions for future work.

7

Utility functions, prices & negotiation

Communicating business intent to self-managing systems

What makes automation easier? a single metric to optimize against

What do business care about? money!

What is money a proxy for? Utility a measure of “goodness”

8John Wilkes, HP Labs: Utility functions, prices, and negotiation, HP Tech Report and Slides, 2008.

Utility In economics, utility is a measure of preferences over

some set of goods and services. Utility represents satisfaction experienced by the

consumer of a good —something a human wants Since one cannot directly measure benefit, satisfaction

or happiness from a good or service, economists measure utility in terms of measurable economic choices

Utility reveals people's willingness to pay different amounts for different goods

9

SLA as Contracts A Service Level Agreement (SLA) is a

contract between mutually suspicious parties if you care about something, put it in the SLA! agreement can be explicit or implicit

Assumptions For self-adaptive systems machine readable, can

be reasoned about Usually involves two-parties: provider, client


WS-Agreement Basics Context

who, why, duration Service terms (specification)

what service is offered, and how it is offered (WSDL) Guarantee terms

Service Level Indicators (SLIs) Observable measures on the service

Service Level Objectives (SLOs) Bounds on client’s and provider’s behaviours SLO violation means that a desired SLO is not met

Price of each outcome How much money changes hands in both directions Penalties and rewards 11

Outcome-based Pricing A better way

Replace all the SLA guarantee terms by a single price function

Specifies how much the service provider is paid for each possible outcome

Omit all details of how the outcomes are achieved


Outcome-based Pricing1-D Price Function


Outcome-based Pricing2-D Price Function


Outcome-based Pricing


Outcome-based PricingPrice Function Only one price function in each SLA priceFunction(set of metrics/parameters) a price

Evaluated by: service provider to determine what to charge client to predict what will happen third party to audit


Outcome-based Pricing Benefits either side can predict price for given outcome can be audited by 3rd party consequences can be explored automatically or

simulated Requirements standalone, deterministic, flexible well-defined, unambiguous, visible inputs

SLIs (Service Level Indicators)


Outcome-based Pricing Pricing is the strategy used for setting prices pricing strategy results in price function

Competition, price pressures sets max prices Customer utility limits what customers will pay demand-elasticity curves


Utility Utility = local measure of goodness more is better!

Arbitrary, local units can be rescaled and re-normalized For example: > 0 win, < 0 lose


Utility for a Fixed Outcome


• Two simple utility functions• Utility for client declines as

the price charge by provider goes up

Utility for a Fixed OutcomeSLIs Service Outcome Price Price Function Utility Functions


Utility for a Fixed Outcome


• Buying a car• Interesting range:

Negotiation range• List price is specified

by the car manufacturer

• Invoice price is what the dealer pays the manufacturer

• Note that utility drops as the price gets too low because of the implied risk of faulty or counterfeit goods.

Utility for a Variable Outcome


• Shaw cable• Telus Internet

Client Utility for two Outcomes


Indifference curves and surfaces

The client is said to be indifferent to (i.e., equally happy with) all outcomes that produce the same value of utility

The curve connecting such a set of points is called an indifference curve.


Combining Outcomes and Price


• Utility as a function of throughput and price

• With indifference curves on the surface representing the function

Minimal Acceptable Utility


Client Utility


Service Provider Utility


Room for Negotiation


Negotiated Price Function


welcome to seng 480b / csc 485a / csc 586a self-adaptive...

Documents