The Zero-ETL Approach: Enhancing Data Agility and Insight
Ultra-large Scale Systems (LSCITS EngD 2011)
1. Ultra Large Scale Systems, York EngD programme, 2010 Slide 1
Ultra Large Scale Systems
Ian Sommerville
2. Ultra Large Scale Systems, York EngD programme, 2010 Slide 2
Objectives
• To discuss the notion of ULSS systems
• To introduce the SEI study report on ULSS
3. Ultra Large Scale Systems, York EngD programme, 2010 Slide 3
Acknowledgements
• Much of the material in this presentation has been derived
from:
• Ultra-large Scale Systems
– Linda Northrop, Software Engineering Institute, Pittsburgh
– Presented at the St Andrews Workshop on Socio-technical
Systems, September 2009.
4. Ultra Large Scale Systems, York EngD programme, 2010 Slide 4
From LSCITS to ULSS
• At around the same time as the LSCITS research and
training initiative was being considered in the UK, a
related study was underway looking at issues around ultra
large scale systems
• LSCITS: 5-10 year perspective
• ULSS: 10-20 year perspective
• ULSS study was about identifying research requirements
rather than doing research.
• Completely independent initiatives but considerable
overlap in conclusions
5. Ultra Large Scale Systems, York EngD programme, 2010 Slide 5
ULSS Report
• ULSS report, published in
2006
• Report of a study
commissioned by US
DoD on the future of
software
• Identifies what is meant
by an ULSS
• Identifies research areas
and challenges
6. Ultra Large Scale Systems, York EngD programme, 2010 Slide 6
Societal problems
• Society is facing major problems and challenges
– Climate change
– Ageing populations in developed countries
– Feeding a growing world population
– Power for a developing world
– Safety and security
– Managing megacities
• Without doubt, these problems can only be addressed using large and
complex software systems
• However, these are ‘wicked problems’ – we cannot specify them in
detail so an evolutionary approach to systems engineering is essential
7. Ultra Large Scale Systems, York EngD programme, 2010 Slide 7
Ultra large scale systems
• A ULS System has unprecedented scale in some of these dimensions:
• lines of code
• amount of data stored, accessed, manipulated, and refined
• number of connections and interdependencies
• number of hardware elements
• number of computational elements
• number of system purposes and user perception of these purposes
• number of routine processes, interactions, and “emergent behaviors”
• number of (overlapping) policy domains and enforceable mechanisms
• number of people involved in some way
8. Ultra Large Scale Systems, York EngD programme, 2010 Slide 8
ULSS
• ULS systems will be interdependent webs of software-
intensive systems, people, policies, cultures, and
economics.
• ULS systems are systems of systems at internet scale
• The boundary between a ULSS and an LSCITS is a very
blurred one. In some cases, it may be appropriate to think
of these terms as synonymous but the general implication
in the report is that ULSS have all of the characteristics of
an LSCITS but at an even larger scale
– The purpose of an ULSS is typically broader than that of an LSCITS
and a ULSS will normally include many LSCITS
9. Ultra Large Scale Systems, York EngD programme, 2010 Slide 9
Do ULSS exist?
• My reading of the report is that it envisages a ULSS as a
web of interconnected automated systems. The ULSS
report assumes that these systems do not yet exist.
• However, what about:
– The World Wide Web
– International air traffic control
– Portfolio of applications in a major multinational company across
several continents
10. Ultra Large Scale Systems, York EngD programme, 2010 Slide 10
ULSS and SoS
• Purpose vs Opportunism
– SoS are integrated and built to fulfil specific purposes
– ULSS have no single well-defined purpose (although they may be
created to support from broad goal such as safe air traffic
management or intercontinental power management). From ULSS,
more specific systems are created to tackle individual problems
• Centralisation vs Decentralisation
– SoS have a single centralised authority and an agreed purpose e.g.
the Navy are the authority for SoS on a warship
– ULSS have no centralised authority and no agreement amongst
stakeholders on the purpose of the system
11. Ultra Large Scale Systems, York EngD programme, 2010 Slide 11
Scale changes everything
• Characteristics of ULS systems arise because of their scale.
– Decentralization
– Inherently conflicting, unknowable, and diverse requirements
– Continuous evolution and deployment
– Heterogeneous, inconsistent, and changing elements
– Erosion of the people/system boundary
– Normal failures
– New paradigms for acquisition and policy
• These characteristics may appear in today’s systems and systems of
systems, but in ULS systems they dominate.
• These characteristics undermine the assumptions that underlie
today’ssoftware engineering approaches.
12. Ultra Large Scale Systems, York EngD programme, 2010 Slide 12
Today’s approaches
• The Engineering Perspective - for large scale software-intensive systems
• largely top-down and plan-driven
• requirements/design/build cycle with standard well-defined processes
• centrally controlled implementation and deployment
• inherent validation and verification
• The Agile Perspective - proven for smaller software projects
• fast cycle/frequent delivery/test driven
• simple designs embracing future change and refactoring
• small teams and retrospective to enable team learning
• tacit knowledge
• Today’s approaches are based on perspectives that fundamentally do not cope with the new characteristics
arising from ultra-large scale.
• The mentality of looking backward doesn’t scale.
13. Ultra Large Scale Systems, York EngD programme, 2010 Slide 13
Systems as buildings
14. Ultra Large Scale Systems, York EngD programme, 2010 Slide 14
ULSS as cities
15. Ultra Large Scale Systems, York EngD programme, 2010 Slide 15
Are ULSS like cities?
• Cities (in developed countries) are controlled entities
– There is an elected local city government that is responsible for
coordinating work on utilities, managing disruption to citizens,
approving plans for development, etc.
– Cities only work successfully because of this local government –
they have not just evolved.
– Without this central control, cities were dangerous places with
high crime levels, dangerous buildings, uncontrolled development,
high levels of disease and poverty, etc.
• Who will play the role of the ‘city council’ for ULSS?
• Will the lack of such a role hinder the development of
ULSS?
16. Ultra Large Scale Systems, York EngD programme, 2010 Slide 16
Socio-technical ecosystems
• Socio-technical ecosystems include
– People, organizations, and technologies at all
levels with significant and often competing
interdependencies.
– There will be competition for resources.
– There will be organizations and participants
responsible for setting policies.
– There will be organizations and participants
responsible for producing ULS systems.
– There will need to be local and global
indicators of health that will trigger necessary
changes in policies and in element and system
behavior.
17. Ultra Large Scale Systems, York EngD programme, 2010 Slide 17
Central challenges of ULSS
• Design and evolution
• Orchestration
• Monitoring and assessment
• Organizational integration
• Regulation and control
18. Ultra Large Scale Systems, York EngD programme, 2010 Slide 18
Design and Evolution
• Specific challenges in ULS system design and evolution stemming directly from
the characteristics of ULS systems:
– Economics and industry structure
– Social activity for constructing computational environments
– Legal issues
– Enforcement mechanisms and processes
– Definition of common services supporting the ULS system
– Rules and regulations
– Agility
– Handling of change
– Integration
– User-controlled evolution
– Computer-supported evolution
– Adaptable structure
– Emergent quality
19. Ultra Large Scale Systems, York EngD programme, 2010 Slide 19
Orchestration and control
• Orchestration and control refers to the set of activities
needed to make the elements of a ULS system work
together in reasonable harmony to ensure continuous
satisfaction of mission objectives.
• Orchestration is needed at all levels of ULS systems and
challenges us to create new ways for
– Online modification
– Maintenance of quality of service while providing necessary
flexibility
– Creation and execution of policies and rules
– Adaptation to users and contexts
– Enabling of user-controlled orchestration
20. Ultra Large Scale Systems, York EngD programme, 2010 Slide 20
Monitoring and assessment
• The effectiveness of ULS system design, operation,
evolution, orchestration, and control has to be evaluated.
• There must be an ability to monitor and assess ULS
system state, behavior, and overall health and well being.
• Challenges include
– Defining indicators
– Understanding why indicators change
– Prioritizing the indicators
– Handling change and imperfect information
– Gauging the human elements
21. Ultra Large Scale Systems, York EngD programme, 2010 Slide 21
Reductionist view of research
22. Ultra Large Scale Systems, York EngD programme, 2010 Slide 22
Post-reductionist research
23. Ultra Large Scale Systems, York EngD programme, 2010 Slide 23
ULSS research areas
• Human interaction
• Computational emergence
• Design
• Computational engineering
• Adaptive system infrastructure
• Adaptable and predictable system quality
• Policy, acquisition and management
24. Ultra Large Scale Systems, York EngD programme, 2010 Slide 24
Conclusions
• There is an unstoppable trend toward increasing scale in many
systems important to our society.
• Scale changes everything. Manifestations of scale and its attendant
complexity arise in many disciplines, and can be understood as a
phenomenon in its own right.
• New, interdisciplinary perspective and new research in building ultra
large-scale systems is long overdue.
• “Since computation has moved over the past twenty years decisively closer
to people, interfaces with social sciences such as Psychology and Sociology,
besides Economics, have become increasingly important”
– Christos H. Padadimitriou, in “Algorithms, Games, and the Internet”
25. Ultra Large Scale Systems, York EngD programme, 2010 Slide 25
Conclusions
• What you call a system (system of systems, ULS system,
complex net-centric system, LSCITS) is really
unimportant.
• It is important that ULS system characteristics are
recognized.
– These characteristics undermine the assumptions we make in most
current technical, management, and acquisition approaches.
– The ULS system perspective is helpful in understanding some of
the current technology and management shortcomings and issues
with system of systems.