CS626 Data Analysis and Simulation Instructor: Peter Kemper R 104A, - - PowerPoint PPT Presentation

1

CS626 Data Analysis and Simulation

Today: Overview & Introduction Instructor: Peter Kemper

R 104A, phone 221-3462, email:kemper@cs.wm.edu Office hours: Monday,Wednesday 2-4 pm

2

Organizational Issues Class materials http://www.cs.wm.edu/~kemper

 PDF files of slides  Homework, projects, …  Supplementary material: tutorials, references, …

Schedule

 MWF: 10.00 – 10.50 am  Office hours: Monday, Wednesday 2pm-4pm and by appointment  No class February 18

• gives extra time for assignments
• due to DSN PC Meeting

Just ask if you are interested what I am doing there, I am happy to tell you more …

3

References

We will use texts from multiple sources! A lot of documents/books are available online in the SWEM library! Data Analysis: Bertholt, Borgelt, Hoeppner, Klawonn, Guide to Intelligent Data Analysis NIST/SEMATECH e-Handbook of Statistical Methods, http://www.itl.nist.gov/div898/ handbook/

• P. Dalgaard, Introductory statistics with R,

Springer 2002, online at SWEM

• B. Everitt, A handbook of statistical analyses

using R, CRC Press 2010, online at SWEM Simulation: Law/Kelton, Simulation Modeling and Analysis, McGrawHill

4

Big Picture: Model-based Analysis of Systems

portion/facet real world formal / computer aided analysis solution, rewards, qualitative and quantitative properties probability model, stochastic process transformation presentation transfer decision description perception solution to real world problem real world problem formal model

5

CS 626 – Focus: Stochastic Models

Stochastic models rely on probability theory

What is probability? A much beloved topic for students A particular type of functions f : S -> [0,1] A mathematical mean to process something we are not sure about:

6

Calculating with something we are not sure about ? What to give as input ?

 Separate “system” from “environment”, clarify on interactions  Separate “subsystems” and their dependencies within “system”  Quantify likelihood of relevant elementary “events” to happen

What to know ?

 How to calculate with probabilities?  How to handle dependencies?

What to gain ?

 Information on likelihood of overall behavior,

quantified information on expected behavior to evaluate a single system or to compare systems

7

Grading: How to get an „A“ for CS 626

Class participation: 0%



Seems useless but is key to meet the other criteria



It will be more fun if you actively participate

 Ask questions, make suggestions, contribute …

Homework: 30%



About 5-6



How to get an A? Just do it, hand in on time, present your results …

Projects: 20%



Will require some effort, time and creativity



Adheres to „learning by doing“ approach



How to get an A? Start early, get things done, hand in on time, reflect what you are doing, present your results …

In-class Exams: 50%



Midterm: 20%



Final: 30%



How to get an A? The usual game …

8

Overview – this is the plan Probability Theory Primer & Statistics Concepts Stochastic Input Modeling

 Different types of stochastic workloads  Relies heavily on data analysis

Simulation Models

 Static  Dynamic

 Discrete Event Dynamic Systems  Continuous, ODE models

Output Analysis

 Data analysis strikes back again!

Verification, Validation, Testing of Simulation Models Data Analysis Classics:

 Preparation,  Finding Patterns, Explanations, and Predictors

Tools:

• Mobius
• BioPEPA
• AnyLogic
• R
• KNIME

9

Probability Theory

has its origins in an interest in games



Cards, roulette, dices,

 Gerolamo Cardano, 1501-1576



From Wikipedia: … notoriously short of money and kept himself solvent by being an accomplished gambler and chess player. His book about games of chance, Liber de ludo aleae, written in the 1560s but published

• nly in 1663 after his death, contains the first systematic treatment of

probability, as well as a section on effective cheating methods.



Pierre-Simon Laplace, 1749-1827

• "It is remarkable that a science which began with the consideration of

games of chance should have become the most important object of human knowledge." Theorie Analytique des Probabilite, 1812.

• Other VIPs of probability theory:

Andrey Kolmogorov Andrey Markov

10

So where to start? With an application example!

My choice: Dependability of a LEO satellite network

Reference

• E. Athanasopoulou, P. Thakker, W.H. Sanders.

Evaluating the dependability of a LEO satellite network for scientific applications. In Proc. 2nd int. Conf. Quantitative Evaluation of Systems, pp 95-104, IEEE, 2005.

What to learn from this:



An impression on what can be achieved with stochastic models



Some terminology, techniques and tools we need to give a closer look

Please keep in mind:



LEO satellite modeling is one application among many



Dependability modeling is one application area among many



The terms and techniques are a subset of what is known

=> there is more to learn here, and it is interesting!

11

A Low Earth Orbit Satellite Network The story:

 University of Illinois student-developed satellite network  Based on Illinois Observing Nanosatellite (ION)  Purpose: collect scientific data

 E.g. natural disaster monitoring, earthquake monitoring, mapping of

Earth’s magnetic field, measuring radiation flux for space weather …

 In particular: measurement of light emissions from oxygen chemistry

in the atmosphere

 Several mission objectives:

 Testing new thrusters, a new processor for small satellites, a new

CMOS camera, demonstration of attitude control on a CubeSat

Some issues from the list of challenges:

 What minimum radiation shielding is necessary ?  What level of redundancy is necessary ?

… to achieve a 6 mth target lifetime with COTS components …

 How does sharing resources through a network could improve

communication with the ground ?

12

Dependability Assessment

Properties of interest, goals of study



Reliability of network R(0,t), probability of no permanent critical system failures during time interval [0,t]



Interval availability A(0,t), fraction of time system delivers proper service during time interval [0,t]

System vs Environment



System:

 Ground station, 45° inclination, northern hemis., repairable failures  4 satellites

 7 critical subsystems (5V/9V regulators, battery, solar panel, comm

hardware, processor, telemetry) + experiment hardware with temp and permanent failures

 Orbits:

 Sat 1-3: 90 min period, Sat 4: 720 min period  Inclination: Sat 1 90° , Sat 2 90° orth, Sat 3 45° , Sat 4 0°



Environment: “the rest” with a foreseen influence based on

 Lightning storms etc cause preamp failures at ground station  Radiation causes failures for satellite processors and CMOS devices

13

Communication

Communication



Satellite - satellite: if within communication range



Satellite - ground station: if within footprint

We discretize orbits, identify matching periods

Intersatellitelink (ISL) Gateway Link (GWL) GWL

Elevation: Angle wrt to center of radiation cone and earth surface

ε

Footprint

14

Clarification: Inclination

Inclination δ δ Satellite orbit Closest point to earth Plane of satellite orbit Equatorial plane

15

More input data

Radiation dose, shielding and its mapping to failure rates,



0.007 failures per year for processor and CMOS components



0.0001 failures per year for discrete components Scale factor r=1 for 1mm shielding for higher orbit Consider several model configurations!

Communication:



Data collection rate: 2 GB/yr while memory available all data lost at failure



Uplink communication considered negligible



Simple routing mechanism



ISL communication rate: 115 kbps with 50% overhead, 226665 MB/yr



GWL rate: double ISL rate



ISL with commercial satellite networks

16

Where are the probabilities ?

Dependability study:



Ground station: rates of failure / repair actions



Satellite subsystems: rates of failure / repair actions are modeled with a random variable that follows a negative exponential distribution with a given rate. Rate 5.0 means on average 5 events per time unit.



Total Ionizing Dose (TID): is taken into account by a scaling factor r towards failure rates of components.

What is then analyzed



Reliability and availability



For different levels of radiation shielding



For different levels of redundancy of components

What type of analysis is used



Transient analysis of Markov chains

 For single satellite design



Discrete event simulation of stochastic models

 valid alternative, used for evaluation of overall network

17

Outcome of calculations Satellite reliability in time interval [0,t] with t in months for different levels of radiation shielding

 Base radiation rate r = 1

(matches 1 mm of shielding)

 Increased shielding: r = 0.4, r = 0.2  Reduced shielding: r = 2, r = 5

Probability of no permanent critical system failures during [0,t]

18

Some more results wrt reliability Different levels of redundancy for

 Batteries

Note, failures are unrelated to radiation Two batteries seem ok

 Regulators

Redundancy does not improve TID immunity since processor or communication system fail due to TID much earlier

19

Results wrt to Performance and Dependability Baseline results for r = 1 A set of simulations experiments are performed for

• Different levels of r
• Different protocols
• Configurations
• Buffer capacities
• Data collection rates
• Communication with
• ther commercial networks

20

What does this study illustrate?

portion/facet real world formal / computer aided analysis solution, rewards, functional properties probability model, stochastic process transformation presentation transfer decision description perception solution to real world problem real world problem formal model

Memo

21

Rene Magritte, The Treachery of Images, 1928-29

What does this study illustrate? Model vs System

 largely simplified formal/mathematical/stochastic model

implemented in software in a fully controlled environment

 set of physical devices interacting in space-time in an largely

uncontrolled, not fully understood environment

Model

 includes some of the rules how the system operates, excludes

• thers

 includes some aspects of the real world as random variables,

ignores others or assumes them as constant

 is parameterized with respect to certain design variables

Study

 has an objective, a clear question  delivers values that are probabilities like R(0,t)

 Interpretation?

 evaluates effects of different design choices

22

23

Open Issues: How does the stochastic model look like ?

 Modeling notation:

 compositional stochastic activity network

 Modeling framework:

 Mobius: multi-paradigm multi-solution framework

from W.H. Sanders, UIUC

How to obtain numerical values used in modeling?

 Data analysis

How do the calculations work ?

 Analysis of continuous time Markov chains  Discrete event simulation of stochastic models

How interpret results and how to decide if computed values are trustworthy?

 Output analysis, Verification and Validation

And what happened with the satellites ?

24

How the story ended:

“A launch attempt with the Illinois Observing Nanosatellite (ION) was made on July 26, 2006. Unfortunately, ION and the 17 other satellites on the launch did not make it into orbit due to a launch failure in the second minute. ION was to be deployed in the fifteenth minute. The launch provider is carefully evaluating the root cause of the failure,and …”

• P. Thakker, http://www.amsat.org/amsat/archive/amsat-bb/200607/msg00444.html

So success is not easily earned with space technology, at least no human was killed in that mission. Obviously a dependability analysis of the carrier rocket system would give us an interesting topic as well.

25

Summary Today:

 Example case study

Future:

Probability Theory Primer, Statistics Concepts Stochastic Input Modeling Different types of stochastic workloads Simulation Models (static, dynamic) Output Analysis Verification, Validation, Testing Data Analysis Classics:

 Preparation,  Finding Patterns, Explanations, and

Predictors