Autonomic computing: the first decade

IBM Research
Autonomic Computing:
The First Decade
Jeff Kephart ([email protected])
IBM Thomas J Watson Research Center
Hawthorne, NY, USA
© 2009 IBM Corporation
IBM Research
Outline
§ Birth
§ Formative Years
§ What Have we Accomplished?
– And what we have not?
2
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
In the beginning there was Chaos
3
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Where it All Began:
The Autonomic Computing Manifesto
§  IBM Senior Research VP Paul Horn
first set forth the idea of Autonomic
Computing in keynote to National
Academy of Engineers
§  Harvard University, October 2001
§  Autonomic Computing Manifesto
released immediately thereafter
4
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
5
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Eight Key Elements of an Autonomic Computing System
6
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
This was soon boiled down to four …
7
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Outline
§ Birth
§ Formative Years
§ What Have we Accomplished?
– And what have we not?
8
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
IBM s Internal Realignment to Support AC
§  Created new Autonomic Computing group within Systems
Management division
– Alan Ganek, VP of Autonomic Computing
– Autonomic Computing architecture board
§  Created a new Autonomic Computing department within
Research Division in 2002
– Approximately 20 individuals
– Approximately 100 researchers working on AC across IBM
§  Created a new Joint Program to guide and fund AC Research
– Dave Kaminsky/Tom Corbi and Jeff Kephart
9
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
IBM wanted to
help drive a new
research agenda
10
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Autonomic Computing Advisory Board
§  We established an AC Advisory Board in 2002
–  Sponsors: IBM Research VPs Alfred Spector, Robert Morris, Tilak Agrawala
–  Chair: Jeff Kephart
§  Mission
–  Help define appropriate research agendas and curricula
–  Contribute insights on what are the relevant problems
–  Stimulate interest in AC issues of relevance within and across their respective fields
–  Endorse and legitimize autonomic computing within industry and academia
§  We recruited 8 top academics and 5 key industry experts
–  Professors of AI, Distributed Systems, Grid Computing
§  We presented IBM s AC research and solicited
–  Feedback on our research
•  More on self-healing, self-protection, human interaction; deeper work on policy; clarify
architecture; build system prototypes
–  Advice on how to enlist academia to work on the great AC challenges
11
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
AC Advisory Board Recommendations for Recruiting Academia
1.  Publish a well-placed, high-quality manifesto
2.  Show that AC is radical, revolutionary, world-changing
a.  Publicize IBM’s own high-quality research in AC
b.  Target top academics
• 
• 
Define problem in their specific terms
If they write good papers, rest of field will follow
3.  Demonstrate industry-wide interest in AC (not just IBM hype)
4.  Organize, sponsor, and participate in workshops, conferences
a.  International conferences and workshops
b.  Special IBM AC workshops
12
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
AC Advisory Board Recommendations for Recruiting Academia
1. Publish a well-placed, highquality manifesto
IEEE Computer
Cover Feature,
January 2003
2. Show that AC is radical,
revolutionary, world-changing
a.  Publicize IBM’s own high-quality
research in AC
b.  Target top academics
•  Define problem in their specific terms
•  If they write good papers, rest of field
will follow
13
ICAC 2011 Keynote
IBM Systems Journal
Vol. 42, No. 1, 2003
Autonomic Computing
June 15, 2011
© 2009 IBM Corporation
IBM Research
Targeting top professors; organizing conferences: Timeline
9 AC FAs
awarded by
Research
2001
AC
Manifesto
University Days: UK,
Supported ~100 faculty-years
9 AC FAs
Germany, France
awarded by
Research Ga Tech
13 FAs,
equipment grant,
2 SURs
2 Toronto CAS
awds
Start
25 profs
15 profs
contacting
supported via
supported via
universities
FA, CAS,
FA, CAS, SUR,
equipment grant, etc.
IIG, etc.
2002
AC Vision
papers (IEEE
Computer, IBM
Sys Journal)
2003
AASMS03 AMS03
(FCRC)
(HPDC)
5 AC workshops
2 AC journals
14
ICAC 2011 Keynote
2004
IJCAI 03 ICAC04
Wkshop
on AC &
AI
26 AC confs/
wkshops
2 AC journals
2005
2006
2006
WRAC05 ICAC06,
HotAC06,
ICAC05 SelfMan05
SelfMan06
Latin American
AC
symposium
30 AC confs/
>40 AC confs/
wkshops
wkshop
3-4 AC journals
>4 AC journals
June 15, 2011
© 2009 IBM Corporation
IBM Research
Organize and sponsor workshops, conferences
AASMS 03
AMS 03
Algorithms & Architectures for Self-Managing
Systems
Active Middleware Workshop on
Autonomic Computing
Federated Computing Research Conference
High Performance Distributed Computing
June 03, San Diego, CA
June 03 Seattle, WA
Chase (Duke), Goldszmidt&Keeton (HP), Kephart&Tetzlaff (IBM)
Hariri (Arizona), Parashar (Rutgers)
ICAC 04
Establish an AC research community
to work together to realize the vision of
large-scale self-managing systems
Develop and nurture the
AC research community
International Conference on
Autonomic Computing
May 17-18, 2004, New York
ICAC 05
10 demos
3 tutorials
June 13-16, 2005, Seattle
ICAC 06
15
ICAC 2011 Keynote
12 demos; 4 tutorials;
workshops
June 12-16, 2006, Dublin
June 15, 2011
5
© 2009 IBM Corporation
IBM Research
AC is catching on!
Excerpt from Report to Paul Horn
on Autonomic Computing as an
Academic Discipline , late 2006
I m
flabbergasted!
§  Initially spurred by our efforts
–  Faculty awards, equipment grants
–  Workshops, conferences (some IBM Academy)
–  University visits
–  Several classes taught by IBMers at Duke, UNC, St
Andrews, Brazil
§  But increasingly on its own
–  AC classes being taught around the world
• 
• 
• 
• 
>30 universities have AC content in their curricula
“Self-Managing Systems”, Shivnath Babu, Duke University
“Autonomic Computing”, Omer F. Rana, Cardiff U., UK, ½ day seminar.
“Parallel and Distributed Computing”, Manish Parashar, Rutgers.
–  Government support: EPSRC in UK funds “Semantic
Grid and Autonomic Computing Programme”
–  Over a dozen AC workshops, conferences initiated by
non-IBMers
–  Publications
• 
• 
• 
IEEE Task Force on Autonomous and Autonomic Systems newsletter
Special Issue of IEEE Internet Computing Jan 2007 on AC
ACM Transactions on Complex Adaptive Systems
–  Web site: www.autonomiccomputing.org
16
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Outline
§ Birth
§ Formative Years
§ What Have we Accomplished?
– And what have we not?
17
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Have we kept the momentum (five years later)?
§  Over 8000 papers on
autonomic computing
–  Approximately 160 ICAC papers
(2% of literature)
§  Over 200 patents issued on
autonomic computing
–  >100 more under evaluation
§  Nearly 200 conferences or
workshops solicit papers on
autonomic computing
§  Government funding
–  FP6: Situated autonomic
communications
•  ANA, BioNETS, CASCADAS,
HAGGLE, ACCA
–  FP7: Self-awareness in
autonomic systems
18
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Let s take a closer look at how AC is doing as a field
§  Run Harzing s Publish
or Perish with queries
Autonomic Computing
and International
Conference on
Autonomic Computing
–  Uses Google Scholar;
finds top 1000 papers in
terms of citation counts
§  Put structured data in
spreadsheet
§  Cleanse the data
§  Identify interesting trends
19
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Autonomic Computing Papers (2002)
417
168
123
84
63
55
52
44
43
42
39
27
24
21
20
15
14
14
12
10
9
8
6
4
4
3
3
2
2
20
D Patterson, A Brown, P Broadwell,
Recovery-oriented
G Candea…
computing (ROC): Motivation, definition, techniques,2002
and case
UC Berkeley
UC Berkeley
TechTech
Report
Report
JP Bigus, DA Schlosnagle,ABLE:
JR Pilgrim…
A toolkit for building multiagent autonomic systems
2002 IBM Systems …
N Zhong, J Liu…
In search of the wisdom web
2002 COMPUTER-LOS ALAMITOSR Sterritt
Towards autonomic computing: effective event management
2002 … Workshop, 2002. Proceedings. 27th An
A LaMarca, W Brunette, D Plantcare:
Koizumi, MAn
Lease…
investigation in practical ubiquitous systems
2002 UbiComp 2002: …
M Satyanarayanan
A catalyst for mobile and ubiquitous computing
2002 Pervasive Computing
D Paulson
Computer system, heal thyself
2002 Computer
GM Lohman…
SMART: Making DB2 (more) autonomic
2002 … of the 28th international conference on
IBMA Computing
IBM's Perspective on the State of Information Technology
2002 White Paper, information available at http:
SS Lightstone, G Lohman…
Toward autonomic computing with DB2 universal database
2002 ACM SIGMOD Record
S Elnaffar, P Martin…
Automatically classifying database workloads
2002 Proceedings of the eleventh …
E Mainsah
Autonomic computing: the next era of computing
2002 Electronics & Communication Engineering
RK Sahoo, M Bae, R Vilalta,
Providing
J Moreira…
persistent and consistent resources through event log analysis
2002
andWorkshop
predictionson
forSelflarge-scale
…
computing syst
CH Crawford…
eModel: addressing the need for a flexible modeling framework in autonomic
2002 Modeling,
computingAnalysis and Simulation of …
DA Patterson
Recovery oriented computing: A new research agenda for a new century
2002 Keynote address, HPCA
WW Gibbs
Autonomic computing
2002 Scientific American
E Schwartz
IBM Offers a Peek at Self-Healing PCS: Autonomic computing initiative2002
will lead
Date
toAlleged:
self-configuring
Nov
desktops and notebo
MN Huhns…
Robust software
2002 Internet Computing, IEEE
D Pescovitz
Helping computers help themselves
2002 Spectrum, IEEE
LD Paulson
IBM begins autonomic-computing project
2002 Computer
Y Tohma
Fault tolerance in autonomic computing environment
2002
A Wolfe
News analysis: IBM sets its sights on autonomic computing
2002 IEEE Spectrum
DJ Clancy
NASA challenges in autonomic computing
2002 Almaden Institute
JY Chung…
„Beyond e-Marketplace & Next Generation e-Business: Grid, Autonomic2002
Computing
4th International
& Web Services
Conference
“
on Electronic
YS Tan, B Topol, V Vellanki…
Implementing service Grids with the service domain toolkit
2002 IBM Corporation
E Grishikashvili, N Badr, DAutonomic
Reilly… computing: A service-oriented framework to support the development
2002 Proceeding
and management
of 3rd … of distributed applic
AZ Spector
Challenges and opportunities in autonomic computing
2002 Proceedings of the 16th international confe
J Kephart
Technology challenges of autonomic computing
2002 OOPSLA
R Sterritt
Towards autonomic computing: effective event management Software Engineering
2002 Proceedings.
Workshop,
27th 2002
Annual NASA Goddard
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Autonomic Computing Papers (2002)
Autonomic Computing Papers (2002)
1000
Citation Count
100
10
1
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Paper Rank
21
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Autonomic Computing Papers (2003)
2595
626
147
108
106
104
104
99
96
96
86
84
84
82
80
78
75
73
73
70
67
61
57
55
53
53
52
47
44
44
42
40
40
38
36
29
28
27
22 27
JO Kephart…
The vision of autonomic computing
Computer
AG Ganek…
The dawning of the autonomic computing era
IBM Systems Journal
H Kreger
Fulfilling the Web services promise
Communications of the ACM
R Sterritt…
Autonomic Computing-a means of achieving dependability?
Engineering of Computer-Based …
J Appavoo, K Hui, CAN Soules…
Enabling autonomic behavior in systems software with hot swapping
IBM systems …
G Kaiser, J Parekh, P Gross…
Kinesthetics extreme: An external infrastructure for monitoring distributed legacy systems… Autonomic Computing …
R Sterritt…
Towards an autonomic computing environment
Database and Expert Systems …
AB Brown…
Undo for operators: Building an undoable e-mail store
Proceedings of the annual conference on …
M Agarwal, V Bhat, H Liu…
Automate: Enabling autonomic applications on the grid
… Computing …
H Cervantes…
Automating service dependency management in a service-oriented component model
Proceedings of CBSE
DM Chess, CC Palmer… Security in an autonomic computing environment
IBM Systems Journal
DF Bantz, C Bisdikian, D Autonomic
Challener…personal computing
IBM Systems …
Y Diao, JL Hellerstein, S Parekh…
Managing web server performance with autotune agents
IBM Systems Journal
R Want, T Pering…
Comparing autonomic and proactive computing
IBM Systems Journal
F Heylighen…
The meaning of self-organization in computing
Information Systems
X Dong, S Hariri, L Xue, HAutonomia:
Chen…
an autonomic computing environment
… Proceedings of the …
A Leff, JT Rayfield…
Service-level agreements and commercial grids
IEEE Internet Computing
V Markl, GM Lohman… LEO: An autonomic query optimizer for DB2
IBM Systems Journal
D Capera
The AMAS theory for complex problem solving based on self-organizing cooperative agents
C Sapuntzakis…
Virtual appliances in the collective: A road to hassle-free computing
Proceedings of the 9th conference on Hot …
P Buhler, JM Vidal…
Adaptive workflow= web services+ agents
… of the International Conference on Web …
A Dan, H Ludwig, G Pacifici
Web service differentiation with service level agreements
White Paper, IBM Corporation
RJT Morris…
The evolution of storage systems
IBM Systems Journal
EM Maximilien…
Agent-based architecture for autonomic web service selection
Workshop on Web Services and Agent-based …
J Jann, LM Browning… Dynamic reconfiguration: Basic building blocks for autonomic computing on IBM pSeries servers
IBM Systems Journal
M Milenkovic, SH Robinson,
Toward
RC Knauerhase…
internet distributed computing
Computer
C Boutilier, R Das, JO Kephart,
Cooperative
G Tesauro…
negotiation in autonomic systems using incremental utility elicitation
… on Uncertainty in …
R Sterritt
Pulse monitoring: extending the health-check for the autonomic GRID
… Informatics, 2003. INDIN 2003. Proceedings. IEE
M Agarwal…
Enabling autonomic compositions in grid environments
JA Redstone, MM Swift…Using computers to diagnose computer problems
… of the 9th Workshop on Hot …
F Berman, G Fox…
Grid computing
JM Deegan…
High reliability memory subsystem using data error correcting code symbol sliced command
USrepowering
Patent App. 10/723,055
T De Wolf…
Towards Autonomic Computing: agent-based modelling, dynamical systems analysis, andIndustrial
decentralised
Informatics,
control 2003. INDIN …
S Elnaffar, W Powley, D Benoit…
Today's DBMSs: How autonomic are they
Database and Expert …
S Hariri, L Xue, H Chen, MAutonomia:
Zhang… an autonomic computing environment
IEEE International …
DM Russell, PP Maglio, RDealing
Dordick…
with ghosts: Managing the user experience of autonomic computing
IBM Systems Journal
G Lanfranchi, PD Peruta, A
Toward
Perrone…
a new landscape of systems management in an autonomic computing environmentIBM Systems …
S Lightstone, B Schiefer, D
Autonomic
Zilio… computing for relational databases: the ten-year vision
… , 2003. INDIN 2003. …
© 2009 IBM
Corporation
Keynote autonomic architecture and its application in e-medicine
June 15, 2011
H Tianfield ICAC 2011 Multi-agent
Intelligent Agent Technology,
2003.
IAT 2003. …
IBM Research
Autonomic Computing Papers (2003)
Autonomic Computing Papers (2003)
10000
Citation Count
1000
100
10
1
1
5
9
13
17
21
25
29
33
37
41
45
49
53
57
61
65
69
73
77
81
85
89
93
97 101 105 109
Paper Rank
23
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
183 papers
Autonomic Computing Papers (2004)
Autonomic Computing Papers (2004)
McKinley, Composing Adaptive Software
10000
Cohen, Correlating instrumentation data to system states: A building block for automated diagnosis and control
Candea, Microreboot—A technique for cheap recovery
Walsh, Utility functions in autonomic systems
1000
White, An architectural approach to autonomic computing
Kephart, An artificial intelligence perspective on autonomic computing policies
Citation Count
Tesauro, A multi-agent systems approach to autonomic computing
Barrett, Field studies of computer system administrators: analysis of system management tools and practice
Müller-Schloer, Organic Computing
100
Parashar, A component based programming framework for autonomic applications
Littman, Reinforcement learning for autonomic network
10
Hellerstein, Challenges in control engineering of computing systems
Ranganathan, Autonomic pervasive computing based on planning
Brown, Benchmarking autonomic capabilities: promises and pitfalls
Kiciman, Discovering correctness constraints for self-management of system configuration
1
1
11
21
31
41
51
61
71
81
91
101
Paper Rank
24
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Total: 183 papers
Autonomic Computing Papers (2004)
ICAC04: 39 papers
ICAC s impact
Autonomic Computing Papers (2004)
10000
McKinley, Composing Adaptive Software
Cohen, Correlating instrumentation data to system states: A building block for automated diagnosis and control
183 papers
Candea, Microreboot—A technique for cheap recovery
Walsh, Utility functions in autonomic systems
1000
White, An architectural approach to autonomic computing
Kephart, An artificial intelligence perspective on autonomic computing policies
Citation Count
Tesauro, A multi-agent systems approach to autonomic computing
Barrett, Field studies of computer system administrators: analysis of system management tools and practice
Müller-Schloer, Organic Computing
100
Parashar, A component based programming framework for autonomic applications
Littman, Reinforcement learning for autonomic network
Hellerstein, Challenges in control engineering of computing systems
10
Ranganathan, Autonomic pervasive computing based on planning
Brown, Benchmarking autonomic capabilities: promises and pitfalls
Kiciman, Discovering correctness constraints for self-management of system configuration
1
1
11
21
31
41
51
61
71
81
91
101
Paper Rank
25
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Wordle s View of AC circa 2003
http://www.wordle.net/show/wrdl/3752036/Autonomic_Computing_paper_themes_2003
26
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Wordle s View of AC circa 2004
http://www.wordle.net/show/wrdl/3752222/utonomic_Computing_paper_themes_2004
27
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Analyzing AC trends over the first decade: a taxonomy
§  Use original AC vision
paper as basis for a
taxonomy of papers
Architecture: Autonomic
elements interact to produce
system-level
self-configuration
self-healing
self-optimization, and
self-protection
§  Engineering challenges
–  Element lifecycle; software engineering
–  Relationships: services, standards, ontology, negotiation
–  System: policy, human interaction, self-*
§  Science challenges
–  Machine learning, optimization & control
–  Understanding and governing emergent system behavior
28
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
AC Paper Trends 2001-2010: Vision, Architecture, etc.
§  Vision and
architecture
papers have
settled to ~10%
–  Appropriate
1
Vision
0.9
–  BAD!
§  Study of
system
properties as a
whole is rising
steadily.
0.8
Vision
0.14
0.12
0.1
0.08
0.6
Arch
Human
System
Vision
Architecture
System
0.06
Arch
Human
System
Vision
0.04
0.5
Human
0.02
0
0.4
2001 2002 2003 2004 2005 2006 2007 2008 20092010
0.3
0.2
0.1
Architecture
Human
System
0
–  GOOD!
29
0.18
0.16
0.7
§  Human
interaction
study, never
prevalent,
became extinct
in 2006
0.2
ICAC 2011 Keynote
2001 2002 2003 2004 2005 2006 2007 2008 20092010
June 15, 2011
© 2009 IBM Corporation
IBM Research
Vision
§  Autonomic Computing
–  Horn, Ganek, Kephart&Chess; Parashar&Hariri; Sterritt
§  Recovery-oriented computing
–  Don’t try to ensure 99.9999% up time for each component
–  Accept that faults are always going to happen; cope with them at system level
–  Micro-rebooting – minimize downtime by designing systems to be quickly rebootable
at multiple levels
–  If it’s fast enough, occasional mistaken reboots are ok
–  Patterson, Fox et al., UC Berkeley
§  Organic and bio-inspired computing
–  Use insights from biological systems to understand and exploit collective behavior
–  KIT, BADS workshop; SASO; Richard Anthony
No work on applying autonomic nervous system principles to autonomic computing !?!
30
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
A tale of two analogies
§  Computer Viruses
– Viruses replicate themselves by co-opting their host’s resources
– Analogies work on several levels
•  Macroscopic: epidemiology, evolutionary trends
•  Microscopic: immune system
– Analogies help us
•  Understand the problem (science)
•  Ameliorate the problem (engineering)
§  Autonomic Computing
– Large-scale computing systems are becoming too complex for humans to
manage. We need self-managing computing systems:
•  Self-configuring, Self-healing, Self-optimizing, Self-protecting
– Autonomic nervous system automatically dilates pupils, increases respiratory
rate, heart beat, etc.
– Analogy to autonomic nervous system helps us describe the effect we want
to achieve
31
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Computer Viruses: Macroscopic Analogy
§  Epidemiology
– Individual = computer
– Social network is important: can
curtail spread relative to
homogeneous mixing
§  Evolutionary trends
– Several great ages of computer
viruses
• 
• 
• 
• 
File infectors
Boot infectors
Macro viruses
Worms
– Heavily influenced by environment
– Co-evolution with host (e.g.
Microsoft Windows)
– Overly virulent viruses are
unsuccessful
32
ICAC 2011 Keynote
1.6
1.4
1.2
1.0
File, Boot and Macro Virus Prevalence
Incidents per 1000 Machines
File
Boot
0.8
0.6
Macro
0.4
0.2
0
1988
1990
1992
June 15, 2011
1994
1996
© 2009 IBM Corporation
IBM Research
Computer Viruses: Immune System
§  Recognize pathogen
–  Unknown: “Innate” immune system combines
“Know thyself” with “Know thine enemy”
–  Known: Vertebrate immune system
specifically detects tell-tale portions
§  Eliminate it
–  Biology: Killer T cells destroy infected host
cell to save host individual
–  Computers: Can often surgically remove
virus from host cell
§  Learn (if previously
unknown)
•  Self/non-Self as proxy for Benign/Harmful
•  Fight self-replication with self-replication
Immune System for Cyberspace
Virus Analyzer
Firewall
Virus Virus
Virus
Petri dish
4
5
Analyze
behavior,
structure
riVsu
8
private
network
1
3
ption
Prescri
Clients
7
View
Extract
signature
Derive
prescription
Virus
2
Administrator
8
6
9
IBM
Widgets Inc.
–  Biology: Each individual does their own
learning; vaccination helps
–  Computers: Learning can be shared
33
ICAC 2011 Keynote
Joe User
Jane Q. Public
June 15, 2011
Gewgaws Ltd.
© 2009 IBM Corporation
IBM Research
How can biological analogies be useful?
§  Marketing: Describe the problem you re
trying to solve; inspire others to solve it
§  Science: Gain insight into the problem
–  Borrow mathematical techniques developed for related
problems
–  Sometimes you end up contributing as much as you
borrow (e.g. directed-graph epidemiology)
§  Engineering: Derive techniques for solving
the problem
–  Knowing that Nature has solved a related problem gives
you hope
–  Even better, you may be able to adapt Nature’s
solutions to your problem
–  Even wrong theories about Nature’s workings can be
valuable!
Immune System for Cyberspace
Virus Analyzer
Firewall
Virus Virus
Virus
Petri dish
4
5
Analyze
behavior,
structure
riVsu
34
ICAC 2011 Keynote
1
3
ption
Prescri
Derive
prescription
Virus
2
Clients
7
View
Extract
signature
Administrator
8
6
9
IBM
Widgets Inc.
Joe User
Jane Q. Public
Be open to what Nature has to teach you,
but be judicious about what ideas you borrow.!
8
private
network
Gewgaws Ltd.
Kephart et al., Fighting Computer
Viruses, Scientific American Nov 1997.
June 15, 2011
© 2009 IBM Corporation
IBM Research
AC Paper Trends 2001-2010:
System architecture, policy, self-optimization
0.2
0.18
0.16
0.14
Self-Optimization
0.12
Policy
Self-O
System
0.1
0.08
System
0.06
0.04
0.02
Policy
0
2001 2002 2003 2004 2005 2006 2007 2008 20092010
35
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Kephart and Walsh, Policy04
How to represent high-level policies?
§  Utility functions map any
possible state of a system
to a scalar value
§  They can be obtained from
Possible
State
σ1
a1
a2
Current
State
S
–  Service Level Agreement
–  preference elicitation
a3
–  simple templates
Possible
State
σ2
Possible
State
σ3
§  They are a very useful
representation for high-level
objectives
–  Value can be transformed and
propagated among agents to
guide system behavior
36
ICAC 2011 Keynote
U(RT)
=
June 15, 2011
© 2009 IBM Corporation
IBM Research
U(RT, RPO)
How to manage with highlevel policies?
§  Elicit utility function U(S) expressed
in terms of service attributes S
U
§  Model how each attribute Si depends
on controls C and observables O
Recovery
Point
Objective
–  Models expressed as S(C; O)
Response Time
–  E.g., RT(routing weights, request rate)
–  Models from experiments, learning,
theory
§  Transform from service utility U to
resource utility U by substitution
Transform
U (cpu, b; λ)
λ=0.01
–  U(S) = U(S(C; O)) = U’(C; O)
§  Optimize resource utility. As
observable O changes, set C to
values that maximize U (C; O)
U
–  C*(O) = argmaxC U’(C; O)
cpu
–  U’*(O) = U’(C*(O); O)
Backup rate b
37
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Unity Data Center Prototype: Experimental setup
Maximize
Total SLA
Revenue
Demand
(HTTP req/sec)
Trade3
Demand
(HTTP req/sec)
5 sec
Trade3
Resource
Arbiter
U(#srv)
U(#srv)
U(#srv)
App
Manager
U(RT)
WebSphere 5.1
Trade3
DB2
Server
Server
App
Manager
App
Manager
U(#srvrs)
WebSphere 5.1
DB2
Batch
Server
Server
Server
Server
Server
U(RT)
Trade3
Server
Chess, Segal, Whalley and White, Unity: Experiences with a Prototype Autonomic Computing System, ICAC 2004
38
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
How App Mgr computes its external resource utility
Alternative to generating
full curve: utility elicitation
Max Utility
Patrascu, Boutilier et al. New
Approaches to Optimization and
Utility Elicitation in Autonomic
Computing, AAAI 2005
U
Resource
Arbiter
Elicit:
(srv)
Number of servers
App
Manager
λ
U(RT)
WebSphere 5.1
Trade3
DB2
Model:
U(RT)
Service-level utility
My controls
Arbiter s controls
Observable
U(RT(C; srv, λ))
Transform: U (C; srv, λ) = U(RT(C; srv, λ))
Internal resourcelevel utility
Optimize:
Optimal internal
control settings
External resourcelevel utility
C*(srv, λ) = argmaxCU (C; srv, λ)
U
(srv, λ) = U (C*(srv, λ); srv, λ)
Chess, Segal, Whalley and White, Unity: Experiences with a Prototype Autonomic Computing System, ICAC 2004
39
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
How the Arbiter determines optimal resource allocation
Decision problem:
Allocate resources
srv* = argmaxsrvΣU i(srvi)
Effectively maximizes ΣUi(Si)
Max Utility
Max Utility
Resource
Arbiter
U 1(srv1)
Number of servers
Number of servers
App
Manager
App
Manager
U(RT)
WebSphere 5.1
WebSphere 5.1
Trade3
DB2
DB2
Server
40
U 2(R2)
Server
Server
ICAC 2011 Keynote
Server
Server
Server
Server
June 15, 2011
U(RT)
Trade3
Server
© 2009 IBM Corporation
IBM Research
Policy and Systems: Status and Future
§  We ve made a good start on
developing the utility-optimization
design pattern
–  Theoretically well-grounded
–  Proven practical in several scenarios
§  But we need to push this work
much further
§  Establish that utility works on a
grand scale in AC systems
–  More than just a few agents and attributes
–  An economy, perhaps?
§  Utility elicitation from humans
Lubin, Kephart, Das and Parkes. Expressive PowerBased Resource Allocation for Data Centers. IJCAI 2009.
(Exploring market-based resource allocation for data centers.)
§  Need planning technologies to
support goal policies
–  More than just an engine
–  Tools for constructing planning domain
descriptions
41
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
AC Paper Trends 2001-2010: Self-*, Benchmarks
§  David
Patterson
warned us that
we needed
benchmarks for
self-{C,H,P} in
order to drive
work in the field
§  It appears that
he was right
Self-Healing
Self-Optimization
0.14
0.12
0.1
Self-C
Self-H
Self-O
Self-P
Benchmark
0.08
0.06
§  We need to
revive the
benchmark
work
Self-Protection
0.04
§  We need more
work on self{C,H,P}
42
0.16
Self-Config
0.02
Benchmarks
0
2001 2002 2003 2004 2005 2006 2007 2008 20092010
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Benchmarks
§  David Patterson noted that
–  Benchmarks drive innovation, but practically
all are performance-related
–  Innovations pertaining to self-{C, H, P} require
appropriate metrics
§  Brown et al. developed benchmarks
for configuration and healing
–  Brown & Keller. A model of configuration
complexity and its application to a change
management system. IM 2005.
Brown & Hellerstein. Benchmarking Autonomic
Capabilities: Promises and Pitfalls. ICAC04
§  McCann et al. recommended metrics
for adaptivity, robustness, autonomy,
sensitivity, stabilization; suggested
adapting existing benchmarks
–  McCann & Huebscher. Evaluation issues in
autonomic computing. GCC 2004
§  Other papers include
–  Consens et al. Goals and benchmarks for
autonomic configuration recommenders. 2005
–  K. Kanoun. Dependability benchmarking for
computer systems. 2008.
43
ICAC 2011 Keynote
After a promising start, work on
autonomic computing benchmarks
appears to have (mostly) stagnated.
June 15, 2011
© 2009 IBM Corporation
IBM Research
AC Paper Trends 2001 – 2010:
Relationships: WebServices/Grid
0.2
0.18
0.16
0.14
Cloud
Web Services
or Grid
0.12
0.1
0.08
0.06
0.04
0.02
0
2001 2002 2003 2004 2005 2006 2007 20082009-2010
44
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Relationships: WebServices/Grid
§  Agent
communication
standards
likely to derive
from services
Article resulted
from
brainstorming
session at Agents
for Autonomic
Computing
workshop, ICAC
2008
§  Foresee
convergence
of autonomic
computing,
web services,
grid interfaces
Brazier, Kephart, Parunak, and Huhns, Internet Computing, June 2009
45
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
AC Paper Trends 2001-2010: AI Technologies
§  Relatively
small but
sustained
effort on AI
technologies
for autonomic
systems
0.08
0.07
0.06
Machine
Learning
0.05
0.04
0.03
Ontology
0.02
0.01
Control
0
2001 2002 2003 2004 2005 2006 2007 2008
2009-2010
46
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Machine Learning
§  Good progress on learning models and
policies
–  I. Cohen et al. Correlating …. OSDI04.
–  G. Jiang et al. Discovering likely invariants of
distributed transaction systems for autonomic
system management. ICAC06
–  G. Tesauro et al. A hybrid …. ICAC06
I. Cohen et al. Correlating instrumentation
data to system states: A building block for
automated diagnosis and control. OSDI04
§  We still need to tackle multi-agent
learning
–  Several interacting learners
–  What are good learning algorithms for
cooperative, competitive systems?
•  Stability and sensitivity characteristics
•  What is sensitivity to perturbations?
–  Opportunities for layered learning
G. Tesauro et al. A hybrid reinforcement learning
approach to autonomic resource allocation. ICAC06
47
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Feedback control
§  Good progress on applying feedback control
to individual autonomic elements
–  Middleware including databases, application
servers: Book and multiple papers by J.
Hellerstein et al.
§  Good progress on applying feedback control
to clusters of compute resources , power
and performance
–  Kusic et al.
§  We still need to understand and control
the behavior of multiple interacting
feedback loops
–  Hierarchical and distributed
–  Some good early thoughts in P.
Ranganathan. No Power Struggles:
Coordinated Multi-Level Power
Management for the Data Center.
–  ASPLOS08
§  Generally, we still need to understand
emergent behavior much better
48
ICAC 2011 Keynote
D. Kusic et al. Power and Performance Management of
Virtualized Computing Environments via Lookahead
Control. ICAC07
June 15, 2011
© 2009 IBM Corporation
IBM Research
Unanticipated trends, and their impact on AC
§  Data centers and energy
management
–  The physical infrastructure is complex,
and needs to be autonomic, too!
–  New attributes: Energy and temperature
§  Cloud Computing
–  Some vendors (Google, Amazon,
Facebook) can get away with highly
standardized and homogeneous
environments
–  Outsourcing to the cloud means that
fewer companies manage IT themselves
–  Perhaps it places a greater burden on
cloud providers to implement AC
•  Lower costs
•  Places premium on easy configurability
•  Outages are more embarrassing and costly
49
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Conclusions
§  Autonomic Computing is alive and well
–  Thousands of papers, 129 of them with at least 50 citations
–  Hundreds of conferences and workshops that touch on AC
§  We have had a busy and fruitful first decade
–  Good balance of vision, architecture, new techniques, apps
–  We haven’t exploited the autonomic nervous system analogy – but that’s OK
–  Not much new theory, or system-level prototypes that address multiple facets
§  Several serious engineering and science challenges remain
–  We need more work at the system level
• 
• 
• 
• 
Multi-agent learning, interacting feedback loops
Understanding/harnessing emergent behavior
Economic models should be pursued seriously
We need to build and experiment with prototypes and testbeds
–  We need to revive our development of benchmarks for Self-{C,H,P}
–  We need more focus on human interaction with autonomic systems; elicitation
50
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
Backup
51
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
ICAC 2004-2011
Year
2004
2005
2006
Location
General Chairs
New York, NY
Jeff Kephart (IBM Research)
Rajarshi Das (IBM)
Manish Parashar (Rutgers)
Vaidy Sunderam (Emory)
Jeff Kephart (IBM Research)
Karsten Schwan (Ga Tech)
Manish Parashar (Rutgers)
Yi-Min Wang (Microsoft Research)
Seattle, WA
Dublin, Ireland Karsten Schwan (Ga Tech)
Yi-Min Wang (Microsoft Research)
2007
2008
2009
2010
Jacksonville, FL Mazin Yousif (Intel)
Chicago, IL
Barcelona
Reston, VA
Program Chairs
Mazin Yousif (Intel)
Omer Rana (Cardiff U.)
Jose Fortes (U. Florida)
Omer Rana (Cardiff U.)
Kumar Goswami (HP Labs)
Jose Fortes (U. Florida)
John Strassner (Motorola)
Kumar Goswami (HP Labs)
Simon Dobson (UCD Dublin)
John Strassner (Motorola)
Manish Parashar (Rutgers)
Simon Dobson (UCD Dublin)
Onn Shehory (IBM Research)
Manish Parashar (Rutgers)
Renato Figueiredo (U. Florida)
Emre Kiciman (Microsoft Research)
2011
52
Karlsruhe,
Germany
Hartmut Schmeck (Karlsruhe, GE)
Joseph Hellerstein (Google)
Tarek Abdelzaher (UIUC)
ICAC 2010 Overview | Jeff Kephart & Canturk Isci
13-Jul-11
IBM Research
ICAC Steering Committee (2011)
§  Jeffrey Kephart, IBM Research (Co-chair)
§  Salim Hariri, University of Arizona (Co-Chair)
§  Manish Parashar, Rutgers University
§  Karsten Schwan, Georgia Tech
§  Emre Kiciman, Microsoft Research
§  Renato Figueiredo, University of Florida
§  John Wilkes, Google
53
ICAC 2010 Overview | Jeff Kephart & Canturk Isci
13-Jul-11
IBM Research
Autonomic computing paper impact
(from Harzing’s Publish or Perish)
Papers:
998
Cites/paper:
30.06
h-index:
75
AWCR:
Citations:
29999
Cites/author:
N/A
g-index:
140
AW-index: 67.04
Years:
11
Papers/author:
N/A
hc-index:
51
AWCRpA:
1881.16
Authors/paper:
2.78
hI-index:
25.45
e-index:
101.85
44
hm-index: 50.95
Cites/year: 2727.18
hI,norm:
4494.42
Query date: 6/12/2011
Hirsch a=5.33, m=6.82
Contemporary ac=6.91
Cites/paper 30.06/11.0/2 (mean/median/mode)
Authors/paper 2.78/3.0/3 (mean/median/mode)
192 paper(s) with 1 author(s)
244 paper(s) with 2 author(s)
248 paper(s) with 3 author(s)
227 paper(s) with 4 author(s)
76 paper(s) with 5 author(s)
11 paper(s) with 6 author(s)
54
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
55
ICAC 2011 Keynote
June 15, 2011
© 2009 IBM Corporation
IBM Research
56
ICAC 2010 Overview | Jeff Kephart & Canturk Isci
13-Jul-11