Challenges to StreamBased Service Intelligence Design ChiHung Chi School of Software Tsinghua Univer

1
Challenges to Stream-Based Service Intelligence
Design Chi-Hung Chi School of
SoftwareTsinghua UniversityEmail
chichihung_at_mail.tsinghua.edu.cn
2
Outline for the Talk

• Needs for Software Service Intelligence
• Difficulties for Software Services Intelligence
  Provisioning
• Intelligence Concept (as Compared to Data and
  Knowledge)
• Data Stream Service Intelligence Platform
• Goals
• Architecture
• Design Issues
• Approach

3

• Needs for Software Services Intelligence

4
Observations

• Ultimate goal of developing software is to make
  all people (or more correctly stakeholders)
  happy (or to have good experience), but how?
• Developer/provider centric in
• Software development
• Software execution
• Software deployment and cost model
• Will this make clients experience better?
• Relationship among software deployment, internet,
  and web. How about business?
• IT (Information Technology) vs. BT (Business
  Tech.)?

5
Shifting of Software to SW Services

• Service over software
• Customized solution to individual problems over
  generic ones
• Example anti-virus software or service?
• Service Intelligence as enabler
• One Wikipedia definition of intelligence
• (from MainStreeam Science on Intelligence",
  which was signed by 52 intelligence researchers
  in 1994)
• A very general mental capability that, among
  other things, involves the ability to reason,
  plan, solve problems, think abstractly,
  comprehend complex ideas, learn quickly and learn
  from experience. It is not merely book learning,
  a narrow academic skill, or test-taking smarts.
  Rather, it reflects a broader and deeper
  capability for comprehending our
  surroundings"catching on", "making sense" of
  things, or "figuring out" what to do

6
Who Needs Service Intelligence?

• Service Providers
• Things on-going in their service provisioning
• Competitiveness in the market
• Requirements for service quality improvement
• Service Requesters / Users
• Realistic service requirements
• Selection for best-fit service providers

7

• Difficulties for Software Services Intelligence
  Provisioning

8
Providers Difficulties in Service Quality
Provisioning

• Dynamics in service quality requirements
• Users expectations
• Competitors offering
• Ambitious users centric in quality demand
• Composition of component services
• Independent
• Partial view
• Subjective user experience over objective
  measurement
• Shifting of intelligence demand from users back
  to providers

9
Requesters Difficulties in Service Quality Demand

• Shifting from technical problem solving to high
  level business demand know less, want more
• Ambiguous/Uncertain service requirements
• Insufficient knowledge on what are possible /
  available in the service provisioning market
  (practical vs. possible)

10

• From Data, Knowledge to Intelligence

11
Intelligence Generation Process
12
Possible Service Intelligence Generation Process
(I)

• Raw Data
• Sources of data
• Monitored / traced data on service instances
  (i.e. behaviors of users and machines / network /
  environments)
• User feedback and rating
• Service description
• Requesters profile
• Both objective and subjective
• Massive in size
• Multiple streams
• Low information entropy level

13
Possible Service Intelligence Generation Process
(II)

• Knowledge
• Understanding of what has happened in previous
  service innovations
• Relationships and dependencies
• Synopsis (summary) and knowledge streams
• High information entropy level
• Intelligence
• Users goal oriented
• Decision making based on collected service
  knowledge streams
• Feedback path
• Adjustment to data collection and knowledge
  streams formation

14

• Data Stream Service Intelligence Platform (DSSIP)

15
Data Stream Service Intelligence Platform (DSSIP)

• Goals
• Stream service intelligence decisions and actions
  based on continuous monitored / provided
  information as stream input.
• Support rapid composition of computational
  intelligence component services
• Self adaptive, automatic system framework with
  respect to the users/environments/providers
  dynamics.
• Allow both continuous and ad hoc intelligence
  queries
• Current focus on six types of intelligence for
  online services
• Requirements
• Trust
• Reputation
• Recommendation
• Security
• System Resource Management
• (Note They are found to share substantial data
  stream services and computational intelligence
  component services)

16
Data Stream Service Intelligence Platform
(DSSIP)

• Architectural View

17
DBMS vs. DSMS

• Persistent relations
• One-time queries
• Random access
• Unbounded disk store
• Only current state matters
• Passive repository
• Relatively low update rate
• No real-time services
• Assume precise data
• Access plan determined by query processor,
  physical DB design

• Transient streams
• Continuous queries
• Sequential access
• Bounded main memory
• History/arrival order critical
• Active stores
• Possibly GB/TB arrival rate
• Real-time requirements
• Data stale/imprecise
• Unpredictable/variable data arrive and
  characteristics

18
Design Considerations and Challenges of DSSIP

• Monitored Data Streams
• Definition for data stream collection
  requirements
• What we actually need vs. what we can possibly do
• Light weight monitoring on service instance level
• Granularity vs. performance
• Selectively on/off data streams on-demand
• Bandwidth reduction, data size reduction vs.
  system complexity
• Dynamic adjustments of data streams granularity
• Zoom-in/-out ability vs. system complexity
• Synchronization among on-demand streams
• Timestamp issue (in, out, )
• Complication with flexible on-demand on/off and
  dynamically adjusted granularity.
• Standardized interface to SS system architecture
• Controls and feedback support

19
Design Considerations and Challenges of DSSIP

• Filtering and Intelligence Knowledge System
• Intelligent, dynamically adjusted filtering
• Definition of on-demand filtering requirements
• Lossy process, how much to lose (No recovery)?
• Transformation of data streams into knowledge
  streams and synopsis as output
• What transformations to support?
• Synopsis, punctuation (assertion about future
  stream contents) are good, but blocking operators
  always exit in computational intelligence
  algorithms
• Tradeoff among approximation, adaptivity, and
  efficiency
• On-demand on/off and dynamic granularity also
  imply prediction on streams requirements
• Constraints
• Limited data/knowledge in memory
• Difficult or too costly to refer archived data
  streams (if available)

20
Design Considerations and Challenges of DSSIP

• Computational Intelligence Function Services (I)
• Definition for knowledge stream services below
• Requirements of knowledge stream services for
  effective intelligence function service support
• Small number N of supported knowledge stream
  services with efficient execution of intelligence
  function services (Note that N minimum?)
• Definition and controls for data stream services
  below
• Models for user behavior and quality of
  experience / satisfaction with respect to
  data/knowledge streams
• Relationship among user behavior, satisfaction
  and quality
• (Contd)

21
Design Considerations and Challenges of DSSIP

• Computational Intelligence Function Services
  (II)
• (Contd)
• Data Stream algorithms
• Single pass, if possible
• Limited memory requirements
• Dynamic granularity support
• Joint between continuous knowledge streams with
  different time sampling rate and static data
• Incremental without blocking operators
• Tradeoff among approximation, adaptivity, and
  efficiency, but with confidence level on possible
  error rate
• Support for N-dimensional vectors with elements
  as open-end ranges
• Distributed computation
• Query language support on knowledge streams

22
Design Considerations and Challenges of DSSIP

• More on Stream Based Computational Intelligence
  Algorithms
• Examples
• Distance between two N-dimensional vectors, each
  attribute of which is a open-end data range.
• Stream versions for traditional computational
  intelligence algorithms such as clustering
• Revisit on these algorithms with new constraints
  from data streams
• Techniques to handle data streams, e.g. random
  samples, sketching, histograms, wavelets, sliding
  windows
• Which ones are good for service intelligence? And
  what adjustments need to be taken?

23
Design Considerations and Challenges of DSSIP

• Intelligence services
• Definition for the requirements and support of
  computational intelligence services
• Definition of component services for effective
  composition of intelligent services above
• Small number of supported component function
  services with efficient execution of composed
  services (Note that N minimum?)
• Composition model
• Quantitative computation of overall quality
  (non-functional) for composite service
• Optimization and tradeoff
• Measurement criteria for service composition as
  well as composite service execution quality
• Effective efficient tradeoff among approximation,
  adaptivity, and efficiency.

24
Design Considerations and Challenges of DSSIP

• DSSIP and SS platform coupling
• Definition of input data streams from SS
  platform to DSSIP with proper control mechanisms
  to turn on/off streams with various sampling
  criteria
• Definition of output intelligence decisions from
  DSSIP to SS platform for actions
• Loosely couple between DSSIP and SS platform,
  with proper collaboration strategies

25
Design Considerations and Challenges of DSSIP

• Pilot Application Domains for DSSIP and SS
  platform
• Pervasive e-health services with personal area
  networks (PAN)
• Video sharing network coupled with China version
  of facebook

26
Design Considerations and Challenges of DSSIP

• Approach Taken
• Start with data stream? (mining approach)
• Start with architecture? (network and system
  approach)
• Start with requirements? (service orientation
  approach)
• This is what we believe and what we start with

27

• Project On-Going with about 40 People
• Questions Please!