Service Oriented Next Generation Networks Modeling_ Pricing and

Document Sample
Service Oriented Next Generation Networks Modeling_ Pricing and Powered By Docstoc
					Service Oriented Next Generation Networks:
    Modeling, Pricing and Optimization

                  Michael Devetsikiotis
           Professor of Electrical & Computer Engineering
                  North Carolina State University

           IEEE Communications Society
          Distinguished Lecturer Tour 2008
Overview and Strategic Trends
•  Main themes:
   –    Services rule!
   –    Ubiquity and mobility
   –    Virtualization
   –    Convergence everywhere: telecom - services - infrastructure
         •  Horizontal integration and fixed-to-wireless convergence
         •  NGN, IMS/SIP services, web services: software/middleware meets
            the telcos
•  Reflected in higher layers: virtual worlds, social
   networking, games, virtual collaboration, tele-presence
•  All enabled by next generation communication networks
•  Big questions:
   –  How to build autonomic service delivery platform(s)?
   –  How to enable end to end, peer to peer, ubiquitous “presence”
      and virtual collaboration?
   –  Effects of new kinds of social network traffic and requirements?
Networking Services the Driving Factors
•  Market and Technology trends:
  –  Growth!! (in $s and numbers)
  –  Diversity of traffic
  –  Convergence
  –  Deregulation and competition
  –  New: QoS, Broadband, ATM, GigE, VDSL, streaming,
     VoIP, VoD, peer-to-peer, and social networks…
  –  We are all in “services”!
•  Challenges and opportunities:
  –  Service innovation amidst competition
  –  Infrastructure design for quality of service
  –  Over-provisioning versus control
  –  “On-demand” or “autonomic” systems
Engineering Motivation
•  Driving forces
   –  Faster computers
   –  Faster lines
   –  Integration of services, multimedia, “convergence”
   –  Service requirements: diversity+mobility+“QoS”
   –  Presence: real and virtual worlds
   –  But need to make profit too!
•  How to deliver reliably and efficiently?
   –  ATM, IntServ, DiffServ, MPLS, Optical, DWDM, NGN
•  Analysis and design not trivial
•  Use quantitative modeling and economics
•  Control: yes, but where?
Services in Networks and Economy
•    Over 70% of advanced economies today in services
•    Components becoming “commodities”
•    Applies to telecom and IT sectors too
•    Services are about “co-production” and “innovation”
•    A new “Service Sciences” discipline is emerging
•    Both human level and software/middleware


       Services/Innovation                        Technology/Resources
           Flexibility                              Congestion, QoS
     Next Gen Nets: Convergence
•  In the new public network: from narrowband to a broadband world
•  Data used to run over a network that was largely built for voice
•  From single media to multimedia
•  From a fixed environment to a mobile environment
•  Convergence: the PSTN, the Internet, wireless, broadcast networks,
    cable TV, are all coming together to service the same sets of traffic and
    to deliver the same types of features and services
•  Convergence occurs in network services, devices, applications,
    industries, humans and machines
•    Major Problems in Information Technology
      –  Change is the only constant
      –  Integration of heterogeneous systems is
•    Definition of SOA
      –  SOA is an architectural style that
          encourages the creation of loosely coupled
          business services
      –  An SOA solution consists of a composite
          set of business services that realize an
          end-to-end business process
•    Loosely Coupled Services
      –  Represent a reusable business function
      –  Remove dependencies on implementation
          specifics through standardized interfaces
      –  Standardized interfaces enable the
          flexibility of SOA
•    Definition
      –  Service-Oriented Networking (SON) an emerging
         network architecture gaining greater overall IT
         efficiency by providing intelligent functionality in the
         network fabric that was previously unavailable or
         impractical to implement.
•    Details
      –  Application awareness in the network fabric is key
      –  Challenges end-to-end principle of networks (“don’t
         touch the payload”)
      –  Assumes that the network can make intelligent
         decisions based on application data
      –  Revisits earlier research in application-aware networks
      –  NGN standards make architecture more flexible
Example: Service Oriented Network
•  SONA shifts view of network from pure traffic
     transport-oriented toward service- and
   application-oriented view.
•    Offload services into the network
      fabric that can leverage
      specialized hardware
      (cryptographic or XML processing
•    In this example, the network offers
      a value added service of securing
      SOAP/XML requests and
      responses inline
•    In certain situations, the network
      could provide a full offload of
      endpoint services (e.g., caching
      stock prices), and would be
      managed by a caching policy
•    Content-based routing typically
      involves applying a rule against
      some part of a service request
      (header or content) to derive a
      token as a result.
•    This token is then used to make a
      routing decision
•    In this example, where requests
      are XML messages, we utilize
      XPath to extract the appropriate
      routing token
•    This value-added service can be
      used to enable service partitioning
      (higher efficiency)
•  Robustness
   –  Admission Control
   –  Load Scheduling
•  Resource Allocation
   –  Concurrency Architectures
•  Security
   –  Concurrency Architectures
•  Performance Optimization
   –  Effectively leverage hardware coprocessors
–  Scalability of the network with network entities

–  Adaptivity of network to changes in state of network

–  Distributed policy-driven dissemination of network
   management data between network nodes

–  Distributed control of the network to connect
   consumers and providers while enforcing appropriate
–  SDP enables a self-optimizing infrastructure that
   optimizes the value derived from IT in a SOA.
–  Our architecture is first of its kind to integrate
   techniques from networking, microeconomics, and
   service-oriented computing to form a fully-distributed
–  The principal component of the SDP is a utility-based
   cooperative service routing protocol which uses
   congestion prices to compute optimal rates and
–  We believe that our service delivery platform is
   applicable to the next-generation of middleware and
   telecommunications architectures.
•    Summary
      –  SON provides exciting new multidisciplinary research
         opportunities in service-oriented computing, hardware,
         software, and networking that could have dramatic effects
         on the development of emerging network services
•    Outlook
      –  Develop a methodology for deciding what value added
         services should reside in the network and where
      –  Given a business process, how can one choose an optimal
         set of network services to leverage given cost,
         performance, SLA constraints
      –  What are the issues with pricing and charging value-added
         services on a commoditized network fabric?
Traffic Modeling

Better models required for performance studies in:
         QoS, admission control, testing

Accurate models of statistical behavior of traffic
   Computationally efficient models of traffic
 Why is Traffic Modeling Important?

•  Need for performance evaluation and capacity planning
•  Accurate performance prediction requires realistic traffic
•  Can use in analysis or testing (“synthetic traffic”)
•  Synthetic traffic matches real traces, is more efficient to
•  Most traffic types in high-speed networks are bursty
•  Burstiness is mainly due to autocorrelation
•  Renewal models assume autocorrelation away for
•  Performance prediction non-realistic without burstiness
History and Motivation
•    Predict performance of switches
•    Telephony: Poisson arrivals and exponential holding
•    Good fit, analytically tractable
•    Cornerstone of telephone network design: “teletraffic”
•    Recall “Math meets the Internet”
•    Laws of large numbers, Palm-Khintchine (superposition)
•    Homogenous, static, predictable (limited variability)
•    Modeling parsimony: average is good enough
•    Later: faxes, telephone modems -> holding times?
•    More recently: Data files, pictures, multimedia…
•    Multiplexing, blocking?
Queueing in NGN Networks



Quality of Service: Delay
Transmission delay:                   Propagation delay:
•  R=bandwidth (bps)                    d = link length

•  L=pkt size (bits)                    s = propagation speed
•  delay = L/R                           (~2x108 m/sec)
             transmission               delay = d / s
A                             propagation


        Queueing delay:

             traffic intensity = La/R
Workload Characterization
Definition: captures source behavior and structural properties of network
    systems, not necessarily restricted to network and link layers.
Performance Measurements:
a) Packet Drop
b) Queuing Delay
c) Jitter at multiplexing points in the network
d) Throughput
Affected by:
1) Variability of streamed real-time VBR Video
 2) Connection Arrival Patterns and their duration
 3) Control actions between the stacks
4) User Behavior that drives network applications
Structural properties and characterization of the WWW (World Wide
    Web) – Static or Dynamic
Global Wired/Wireless Internet that impact network performance
Open Research Problems on workload
Multifractal Traffic Characterization
•  Ethernet data from Belcore – Poisson connection arrivals with
   heavy-tailed connection duration times lead to self-similar burstiness
   in multiplexed network traffic
•  TCP Connection arrivals exhibit self-similarity
•  WAN IP traffic has revealed multifractal structure in the form of
   nonuniform scaling across short and long time scales.

Synthetic Workload Generation

Workload Monitoring and Measurement
Open Research Problems on workload
characterization (cont’d)
Spatial Workload Characterization
•  Mobility Model – Understanding the movement pattern of mobiles is
    relevant for effective resource management and performance
    evaluation (random walk, Poisson number of base stations,
    exponential stay durations etc.). It would be of great interest to find
    out a correlation structure at large time and/or space scales (e.g.
    students on campus move from class to class at regular intervals
•  Logical Information Access Pattern – Based on the
    Socioeconomic fabric of everyday life, it becomes relevant to
    characterize the information access pattern my information content.
•  User Behavior – Most network application are driven by users (e.g.
    interaction of a web browser GUI etc. It can be a function of time-of
   -day, pricing, network congestion, response time. Also some users
    can interact and take decisions cooperatively or selfishly leading to a
    noncooperative network.
Performance Analysis
There are numerous challenges on the performance analysis since the
    resource dimensioning using buffer sizing is an ineffective policy.
•  The queuing results are asymptotic in nature where buffer capacity –
    in some form – is taken to infinity to achieve tractability. Little is
    known about finite buffer systems except for observations on the
    dependence of packet loss rate on the “effective” time scale induced
    by buffer size.
•  In the modern network environment with multimedia and other QoS
   -sensitive traffic streams comprising a growing fraction of network
    traffic, second order performance measures in the form of jitter such
    as delay variation and packet loss variation are of importance to
    provisioning user specified QoS.
•  Self-similar and Long tail workloads can make difficult to find the
    equilibrium since they have slow convergence properties.
Open Research Problem in
Performance Analysis

•  Second Order Performance Measures: Impact of
    self-similarity on jitter.
•  Persistent periods of high and low contention
    implied by self-similar burstiness can exert
    negative effect on second order performance
•  Short-range versus long-range correlation
•  Impact of packet scheduling
Traffic Control Open Problems
Especially in broadband wide area networks where the
   delay-bandwidth product is especially severe, and
   mitigating the performance degradation due to outdated
   feedback is critical to facilitating scalable, adaptive traffic
•  Multiple Time scale traffic control
•  Multilayered Feedback control
•  Workload sensitive traffic control
•  Optimal Prediction of long-range correlation structure
•  Dynamic Admission control
•  Optimistic traffic control for short-lived connections
Traffic Analysis Methodology
•  Decompose into scales and “modulated
•  Measure then statistically fit, then validate via
   goodness-of-fit (recall simulation lectures, input
•  Synthetic generation is step for simulation
•  Otherwise analyze with queueing (exact or
Real Traffic: Scales and Burstiness
•  Measurements say traffic is not Poisson
•  Variable with different time scales
Traffic Bursts and Scales

•  Hierarchical view (figure from
   I. Kaj):
   –  Call
   –  Burst
   –  Packet
•  Also “transactions” and
•  How to characterize: Mean?
   Peak? Variance?

          Low Correlation

          High Correlation
 Taxonomy of Models

•  See Frost/Melamed, Adas, Devetsikiotis/Fonseca articles
•  DES, continuous inter-arrivals vs. discrete time
•  Workload, burstiness
   –    Renewal and IID: no dependence
   –    Phase renewal
   –    Markov and embedded Markov: one step memory
   –    Markov modulated, On Off, Interrupted Poisson Process etc.
   –    Markov renewal modulated
   –    Semi Markov
   –    MAP
•  Other interesting models:
   –  Fluid: useful for analysis and also for simulation
   –  Regression models – classical statistics
   –  Discrete autoregressive (DAR) and modular autoregressive (TES)
 Regression Models

•  Autoregressive models

•  TES: Transform-Expand-Sample
   –  Uses modulo-1 operations
   –  Correlated numbers with desired
   –  Marginal by inverse transform
   –  Numerical fitting of correlation

•  DAR(p), for example p=1:
Taxonomy of Applications
•    Hierarchy of scales and components
•    From simple to complex
•    Marginal distributions and correlations
•    From renewal and memoryless to very correlated
     and “long range dependent”
Voice: The mother of all models
•    Simple traffic
•    Two states: active and silent
•    Basic assumptions hold well
•    Models
     –  Semi-Markov (periodic arrivals in Markov states)
     –  MMPP (Poisson arrivals in Markov states)
     –  Fluid model (constant fluid during Markov states)
ON-OFF Source
•  Alternating periods of silence and of activities
•  Voice source:
   –  ON    0.4 – 1.2 sec
   –  OFF   0.6 – 1.8 sec
   –  170 ATM cells/sec in ON
TCP/UDP Traffic
•  TCP and related applications
   –  FTP
   –  TELNET
   –  SMTP
•  Empirical distributions and inversion
•  Interactive and connection setup aspects
•  “tcplib” library: hierarchical model based on
    functionality (see paper)
HTTP and WWW Traffic
•  First analyzed by Deng (see paper)
•  Hierarchical decomposition:
   –    WWW request arrivals
   –    ON periods (data activity)
   –    OFF periods (thinking times, etc.)
   –    Distribution during ON period
•  Measure and fit
•  Assumptions!
•  Heavy-tailed distributions: Weibull and Pareto
Earlier Game Traffic
•  Measurement-based models
•  Hierarchical fitting
•  Example in paper by Borella: micro-scale
A Real-Life Hierarchical Model
•  ETSI model for UMTS (3G) testing
 Video: The most bursty traffic!
•  Compressed video (MPEG)
•  Uses intra and inter-frame compression
•  Based on “frames” and groups-of-pictures (GOP)
Video Modeling

•  MMPP and Markov Modulated fluid models (see
    Schwartz textbook)
•  Autoregressive models
•  DAR(1) and TES
•  They fit adequately but queuing is imprecise
•  In reality, “long range dependent”…
IP-TV Services: they can be classified by their type of content and
•  On demand content: With pre-encoded and compressed content, a
    customer is allowed to browse an online movie catalogue, to watch
    trailers and to select a movie of interest.
•  Live content: a customer is required to access a particular channel
    for the content at a specific time, similar to accessing conventional
    TV. The live content over IPTV can be showed of a live event or a
    show encoded in real-time.
•  Managed services: Video content can be offered by the phone
    companies who operate IPTV business or obtained from syndicated
    content providers, in which the content is usually well-managed in
    terms of coding and playout quality.
•  Unmanaged Services: accessing any third party over the internet
    such as Youtube, Gooble Video etc. However there is not any
    guarantee of playout quality and performance.
 DVB-H/DVB-T & IP-TV Transported Channels
Digital Video Broadcasting (Handheld/Terrestrial) – Some studies show that
    channel popularity distribution (probability of channel being watched by an
    active user) has a power-law from (Zipf distribution)

•    Assuming that the probability πκ that a channel is watched is given by the
      formula: 1 / d kα
•    For k=1,2,…,K which is the channel’s popularity rank, d is normalization
      constant, and K is the number of channels. As α increases, the median of
      the Zipf distribution shifts to smaller k and the distribution becomes less
     -tailed. For example from [15] we can have
Future Trends of Video Modeling
•  High Definition Video (HD): Modeling of H.264
   can be done by a Discrete-Time Semi-Markov
•  Future Trends: Future IPTV expected to provide
   high quality video (HDTV from 4Mbps up to
   13Mbps a channel), mobility such as mobile TV
   (since there is a proliferation of multimedia
   mobile devices such as iPhone, iPod,
   BlackBerry, PDAs and cell phones) and support
   of IP based triple play or quadruple services in a
   single device.
New Era: Self Similarity and LRD
•  Observations in early nineties:
    –  Scaling in Ethernet traffic
    –  Persistence of video traffic
    –  Similarly for WAN
•  Traffic did not scale or become smoother as expected
•  Poisson assumption questioned
•  Birth of “self similar” era
                                                                    Long Range Dependence
 1 Second Intervals     10 Second Intervals   60 Second Intervals

                                                                           Hyperbolic decay of

                                                                       r (k ) ≈ H (2 H − 1)k 2 H − 2   k →∞

 100 Second Intervals   Half Hour Intervals   1 Hour Intervals

                                                                       ∑ r (k ) = ∞       if       1 〈H 〈1
                                                                    k = −∞
 Peer2peer Networks (e.g., eDonkey,
      emule, Gnutella, Kazaa)
•  Peer-to-peer file sharing applications
   have evolved to one of the major traffic
   sources in the Internet.
•  In particular, the eDonkey file sharing
   system and its derivatives are causing
   high amounts of traffic volume in
   today's networks.
•  The eDonkey system is typically used
   for exchanging very large files like
   audio/video CDs or even DVD images.
•  Peer2Peer Network based on
   “torrents”: function different than other
   P2P programs. A large file is divided
   into chunks. Peers interested in the
   same file self-organize into a torrent
   and peers exchange file chunks with
   each other.
Modeling Game Traffic
•  In order for a game with client-server topology to be effective it must
    accommodate huge lag (ping, roundtrip delay) and loss
•  Methods to solve the lag in games:
1.  Client-side prediction of the game state i.e. movements of objects
    and other players
2.  Combine movement with inertia or reducing maximum velocity of
    objects prediction is even more effective
The most robust games tolerate lag up to 1 sec (mean value is usually
    around 200ms) and loss up to 40%

Quality based on Ping times
  <50ms                   Excellent
  <100ms                  Good
  >150ms                  Bad
Generally is based on the robustness and type of the game (there are
  games that accommodate lag of 200ms)
Modeling Games an Open Issue
•  Games have different requirements (a 3D first
   shooter game requires very low, but 2D strategy
   games can accommodate higher lag)
•  Clients do not act independently (social
   networking aspects)
•  Server traffic per client is dependent on all clients
   (many models assume independency)
•  Each game sends different messages and there
   is not currently a general social formalization.
•  Will 3D Virtual Worlds become the “Web 3.0”?
  How to measure traffic?
  Collection of traffic statistics is currently performed
  •  Flow monitor e.g. Cisco NetFlow
  •  Sophisticated Network Monitoring Equipment
  The port based measuring system (identify applications
     based on ports) captures only 70% of traffic
  Solution: Traffic measurements based on service flows
     (looking IP headers)

T. Karagainnis, K. Papagiannaki, and M. Faloutsos, “BLINC: Multi-level Traffic Classification in the Dark”, in Prof. of ACM SIGCOMM,
                                                            Aug. 2005
Overview of Work with IBM
•  Service Oriented Networking and pricing

•  IMS, SIP and telco infrastructure and brokering

•  Large scale service delivery and network service

•  On-demand software performance lab
    –    Monitoring for performance
    –    Rare-event estimation for services, modeling
    –    ITM for VCL
    –    Statistical analysis and resource scheduling
Open Source Software Collaboration
•  Virtual collaboration in software development
•  Received IBM Jazz Faculty Award together with
   Laurie Williams
•  Started demos in CSC 326 and plan to introduce
   labs and tutorials
•  Exploring combinations of Eclipse, Sangam and
•  Looking to embed into 3D virtual interface based
   on Croquet
•  Studying student interactions for software
   development as real life trial
Nortel SIP and Next Gen Services over Mesh Wireless

 •  Partnering with Carleton University in Ottawa,
 •  Analysis: Cross-layer modeling of performance
 •  Trial: Wireless mesh testbed in EB-II
 •  Would benefit from Centennial campus-wide
    wireless network (as originally planned)
ArtCity: Network-Enabled Art
•  Autonomic service delivery platform for the Arts
•  Enabling artistic virtual organizations and remote interactions by use
   of high speed networking and on-demand service delivery.
•  Combine network services with virtual collaboration research, and
   with hands-on, “living lab” setting on campus (immersive Art Village
   in dorm, Centennial trials and pilot event in EBII).
•  Use Centaur lab as hub for connectivity.
•  RENCI and other telepresence and mixed reality facilities (e.g.,
•  Use-cases: wireless-based mobile gaming and virtualized dance
   activities: also serve as sources of system performance and
   workload measurements and analysis.
•  Measurement phase followed by a design phase, where the
   algorithms and protocols in Nortel-sponsored wireless mesh trial can
   be adapted for optimized performance in real-life setting.
•  Our work on autonomic service-delivery platforms and game-based
   resource allocation will be tested and tried in this environment and its
   performance will be tuned accordingly.
 Cisco Broadband Aggregation Architectures
1.  Building emerging traffic and usage patterns based on social networking patterns
2.  Capturing the basic quantitative design parameters of the network architecture with respect to the
    degree of centralization and clustering
3.  Business aspects such as cost models, availability requirements, customer utility and pricing

                                     Centennial Campus
  •    Generates Next Generation Traffic (multi-disciplinary R&D, government facilities, interactive
  •    Idea of building a Wireless Infrastructure Networks (Wi-Fi/WiMAX) and study the MMORPG
       (Massively multiplayer online role-playing game) traffic pattern
  •    Use of Centaur Lab as a gateway to internet 2 and implement traffic monitoring (Linux Boxes, Cisco
       IOS Netflow)

                              NC Centennial Living Labs
Virtual World Trials for EOL and VCL
  •    Engineering-on-Line supporting TA to study Virtual Worlds for ECE 776
  •    Comparison of Virtual Worlds: Second Life, Protosphere, and Qwaq as “3D-
       I” or “Web 3.0” media for remote instruction
  •    Collaborating with College of Management and their similar trials together
       with Indiana U.
  •    Transforming VCL from a static access Lab to a collaborative tool
  •    Looking at Croquet as the long term open source unifying platform

             Our View towards the evolution of the Internet
Web 1.0         Interaction      Web 2.0         Collaboration          Web 3.0
                                 Blogs & Wikis              Social Networking & Co-Creation
       3D Synchronous e-learning
E – learning is an
interactive process in
which student can choose
between                    Notion of interactivity (I)
•  Voice                              and
•  Chat                         Immersion (I)
•  Text                           to achieve
•  White Boards                Engagement (E)
•  Breakout Rooms
•  Application Sharing             *Dr. KARL TAPP

3D Virtual Worlds
3D Synchronous e-learning (cont’d)

•  Shared Experience and Shared Learning
•  Co-creation and Collaboration
•  Social Environment
•  Innovation and Simulation
•  Incentives
•  Informal Learning
•  Use of Avatars
           Limitations of 3D virtual worlds

•    Client - Server model (Network-Computation-Resources)
•    Commercial Products
•    Licenses do not remain on the creator
•    Little end-user flexibility
•    Do not allow to share *ALL* applications




 Examples: Second Life, Forterra Inc, Protosphere etc
 The Croquet Based Platform
•    Peer-to-Peer Architecture
•    Computation of VW is done locally
•    Open Source
•    Integrate any application
•    Strong candidate for Web 3.0
                                         Optional Use of Server

                          P2P network
“The Virtual Computing Lab (VCL) is a remote access service
that allows to reserve a computer with a desired set of applications
for yourself, and remotely access it over the Internet”

 •  Users have remote desktop
 access to machines loaded,
 on demand, with the desired
 •  Anytime-anywhere access
 to applications, transparent to
 •  Ease of system
 configuration and
 management, and scalability.

         Does not support collaboration among users, yet!
                             VCL 3.0

•  Web 3.0 will provide access to multiple knowledge partners and
    diverse types of data that enable on demand assembly of the
   necessary resources and expertise as needed.
•  Social structures in a Virtual Collaboration environment are not static
Qwaq Application Sharing and Remote Desktop
•  For our trials we used both Qwaq Forums (as the license-oriented
   version of Croquet) and Cobalt/Open Source Croquet

Sharing my
Desktop on
the VW

inside the
Croquet Application Sharing and Remote Desktop

 Editing a text file

Sharing a Linux machine

Importing any application
Serious Gaming Enabled by Networks
•  NIH proposal: efficacy of an Internet-based gaming intervention to
   influence physical activity and healthy lifestyle behaviors in the
   college-age population
    –  Despite national recommendations, the incidence of obesity and chronic diseases
       associated with physical inactivity continues to increase
    –  Create a 3D virtual world prototype that will immerse college students in fitness
       and health-related learning
    –  Create a game-based intervention for the physical activity village in the virtual
       world. The design of our game – an “alternate reality game
    –  Evaluate the effectiveness of the virtual world ARG in influencing students’
       physical activity and health-related outcomes
    –  Evaluate the psychological attractiveness of the virtual world ARG as an
•  NSF Proposal: Presence Aware Networking
    –  Mobile HRG to study location and presence awareness through the
       layers, emphasis on networking solutions
Network-Enabled Collaboration for Innovation
Enabling                                         VIRTUAL ORGANIZATIONS
Mechanisms                                                                                                                                              Partners:

                                                                                                                          Open Source S/W (Jazz, VCL)
                 Virtual Public Schools (K-12)

                                                                                                     Medical Technology
                                                                 SSME Community

                                                                                  Serious Gaming
                                                 Art City


                                 Social Innovation                                            Industry/University
                                                                                             Commercial Innovation
Proximity:                                                  Centennial Living Labs
Testing &
Implementation                                                    Virtual RTP
 Our Research in Network Services
•  Service networking and optimization
   –  Service delivery pricing and optimization
   –  Service-oriented networking
   –  Autonomic platform for service brokering and
   –  Measurement-based control of service centers/
   –  Virtualized server characterization and control
•  Networked virtual collaboration
•  Cross layer and wireless design
   –  WiFi and WiMax QoS modeling
   –  Cross layer modeling, simulation, optimization
   –  Mesh and multihop systems (WiFi, WiMax,
Network Service Research Methodologies
•  Pricing, utility-based and game-theoretic resource
•  Stochastic workload and traffic characterization
   for “social networks”
•  Advanced sampling and simulation techniques
•  Cross-layer response surfaces, meta-modeling,
•  Experimental modeling and estimation
   techniques: in vitro and testbed trials
 Some Highlights
•  Leadership in networking and virtual collaboration research and
•  Strong interactions with industry, past and present
    –  IBM       -- Alcatel                 -- Nortel
    –  Tekelec   -- Ericsson IPI   -- EMC
    –  Cisco
•  IBM projects: appliances and the “virtual computing lab” (VCL);
   communication in virtual teams; virtual collaboration in software
   development (Jazz and Rational faculty awards)
•  Nortel: next gen services and presence over wireless mesh testbed;
   service acceptance modeling
•  Cisco: effects of “social networking” on access and router architecture
•  Alcatel: network “self-*” or “self-X” (similar to “autonomic”)
•    Generalized Service Model
      –    Service classes: Hard real-time, delay-adaptive, rate-adaptive and elastic

•    Proposed Generalized Framework
      –    Reserved resource, Cr, for hard-real time services
      –    Shared resource, C’, for the rest of applications
      –    Achieved by Hierarchical Fair Queueing scheduling or similar
•    Measurement Module
      –  Accurate input estimation
      –  Queue monitoring for delay consideration and congestion avoidance
•    Decision Module
      –  Optimal allocation for under and over-provisioned cases
•    Scheduler (e.g., WRR, WFQ)
      –  Update the scheduling weights to realize adaptive resource control
•    Income or Utility Component
      –  Maximize utility charge
•    Cost Component
      –  Minimize delay-incurred cost
•  Originally formulated for network access and edge nodes
•  Adapted to generalized “service nodes”
•  Discussion on Service Guarantee Constraints
    –  Loss
         •  Generalized Rate: Mean/Peak/Effective Bandwidth
    –  Delay
         •  Average/Stochastic/Deterministic Delay Bounds
•  Discussion on Pricing Models
    –  Linear
    –  Nonlinear
•  Scheduling Algorithms
   –  Static WRR and WFQ,
      Dynamic WRR and WFQ
      and our Proposed Optimal
      Dynamic WFQ (ODWFQ)
   –  Our proposed ODWFQ
      achieves the best
      performance in loss and
      delay and stability as the
      narrow confidence interval
•  Motivation
    –  Extend to generalized n classes
    –  Limitations of the previous solution approach under linear pricing models
        •  Scalability Problem: O(2n)
        •  QoS constraints complicate the solution procedure
        •  Over-provisioned vs. under-provisioned
•  Objectives:
    –  Propose a new solution strategy to tackle scalability problem
    –  Propose a nonlinear pricing model that takes care of QoS considerations
       on its own
Nonlinear Pricing Model – I (cont’d)
•  Revenue Per Service i       r (φi ) = pi ⋅ φi

•  Cost function      c(ϕi ) = bi ⋅ exp  βi (D(ϕi ) − di )
                                                          
bi reimbursement price for service i when SLAs violated
Di(φi): performance metric
di: metric threshold
β: steepness of the cost function
Stochastic Bounds
Supplier’s Utility Function
Convex Optimization Problem
•  Model

    where βi > 0 and 0 < αi ≤ 1
    –  αi is a proactive factor and βi is an aggressive factor
•  Two-tier delay differentiation
    –  Differentiation among the classes: bi
    –  Differentiation inside the class:
        •  Sensitive to the desired delay of the class, di