design and run time quality of service management
play

Design and Run-Time Quality of Service Management Techniques for - PowerPoint PPT Presentation

Design and Run-Time Quality of Service Management Techniques for Publish/ Subscribe Distributed Real-Time and E mbedded Systems http://www.dre.vanderbilt.edu/~jhoffert/dissertation.pdf Joe Hoffert jhoffert@dre.vanderbilt.edu Institute for


  1. Design and Run-Time Quality of Service Management Techniques for Publish/ Subscribe Distributed Real-Time and E mbedded Systems http://www.dre.vanderbilt.edu/~jhoffert/dissertation.pdf Joe Hoffert jhoffert@dre.vanderbilt.edu Institute for Software Integrated Systems Vanderbilt University Nashville, Tennessee

  2. Context: QoS-enabled Publish/Subscribe for DRE Systems • Pub/Sub enables separation of concerns - decouples senders & receivers • QoS enables finer-grained control of system behavior/properties Client-server technology may not suffice for all DRE systems => Application Application move towards publish/subscribe ‘Global’ middleware Application Data Store • i.e. , client-server & pub/sub Application are complementary Application technologies Characteristics of Pub/Sub WS Brokered Notification Data Distribution Service • Decouples location via anonymous Java Event & pub/sub Message Notification • Decouples time via asynchronous, Service Services time-independent data distribution Manifesto for QoS-enabled Pub/Sub • Decouples redundancy via The right data…to the right place…at unbounded # of senders/receivers the right time 3

  3. Example: QoS-enabled Pub/Sub DRE Systems • Net-centric & large-scale “systems of … systems” … • e.g., satellite systems, shipboard computing environments, emergency response systems • Satisfying tradeoffs between multiple vs. (often conflicting) QoS demands vs. • e.g., security, timeliness, reliability • Regulating & adapting to vs. (dis)continuous changes in runtime environments • e.g., online prognostics, dependable upgrades, availability of critical tasks, dynamic resource management DRE systems increasingly realized by composing loosely-coupled services ( e.g. , pub/sub) 4

  4. Challenges in Realizing DRE Pub/Sub Systems Variability in the solution space (both design- and run-time) • Diversity in platforms, languages, protocols & tool environments • Enormous accidental & inherent complexities • Continuous evolution & change Data priority • Management of diverse QoS requirements Data reliability Inter-arrival data Data spacing redundancy Ordered data Determining liveness Data for late arriving Data with time readers deadlines Provisioning of data resources Focus on QoS Management 5

  5. Overview of QoS Management Focus Areas QoS Mechanism 2 (adequate) QoS QoS Mechanism 1 Mechanism 1 (adequate) (inadequate) (initial environment) (environment … modification) System Execution Timeline My PhD dissertation addresses 4 aspects of QoS management complexity. 1. Developed model-based techniques to reduce manual effort & ameliorate accidental complexities in deploying pub/sub DRE systems. 2. Designed new composite metrics & a flexible middleware framework to evaluate & benchmark QoS mechanisms. 3. Designed machine learning-based adaptation logic to provide accurate configurations & predictable response times in flexible envs. 4. Designed monitoring mechanisms & improved machine learning- based logic to improve adaptation accuracy in dynamic envs. 8

  6. QoS Management Focus Areas Overview (cont.) QoS Mechanism 2 (adequate) QoS QoS Mechanism 1 Mechanism 1 (adequate) (inadequate) (initial environment) (environment … modification) System Execution Timeline 1. QoS Configuration Development Support: QoS configurations can have numerous entities & QoS policies; how can we help DRE developers manage the complexity of developing configurations? 9

  7. QoS Management Focus Areas Overview (cont.) QoS Mechanism 2 (adequate) QoS QoS Mechanism 1 Mechanism 1 (adequate) (inadequate) (initial environment) (environment … modification) System Execution Timeline 2. Evaluation of QoS mechanisms: Several QoS mechanisms are available; how can we MuxModule FLEXMAT help developers evaluate MulticastModule Testbed QoS mechanisms for … pub/sub middleware? SequencerModule ANT Framework 10

  8. QoS Management Focus Areas Overview (cont.) QoS Mechanism 2 (adequate) QoS QoS Mechanism 1 Mechanism 1 (adequate) (inadequate) (initial environment) (environment … modification) System Execution Timeline 3. QoS Configuration for cloud computing environments: Cloud computing resources which affect QoS aren’t known until runtime; how can we Multicast TCP/IP configure the middleware Custom UDP/IP based on resources provided? protocol 11

  9. QoS Management Focus Areas Overview (cont.) QoS Mechanism 2 (adequate) QoS QoS Mechanism 1 Mechanism 1 (adequate) (inadequate) (initial environment) (environment … modification) System Execution Timeline 4. QoS adaptation in dynamic environments: QoS Mechanism 2 (adequate) As environments or operating QoS Mechanism 1 QoS Mechanism 1 conditions change, QoS can (inadequate) (adequate) diminish; how can we adapt the middleware to support (initial environment) (environment modification) predictable QoS? System Execution Timeline 12

  10. QoS Management Focus Areas Overview (cont.) QoS Mechanism 2 (adequate) QoS QoS Mechanism 1 Mechanism 1 (adequate) (inadequate) (initial environment) (environment … modification) System Execution Timeline Presented solutions to these in qualifying exam; briefly review here Manual QoS Configuration QoS Configuration Focus Area 1 Techniques Validation (design) System Deployment System Lifecycle Timeline QoS? QoS? App App Focus Area 2 ‘Global’ QoS? Protocol 1 Protocol 2 App Data QoS? App Store QoS? Operating App environment 13

  11. Distributed QoS Modeling Language (DQML) Focus Area 1 Focus Area 2 Focus Area 3 Focus Area 4 X 10ms 15ms • Associations • Parameter types • QoS policy • Parameter values parameters Application- specific interpreter DQML addresses the QoS configuration management challenges of DQML addresses the QoS configuration management challenges of DQML addresses the QoS configuration management challenges of (1) Correctly specified QoS properties, (1) Correctly specified QoS properties, (1) Correctly specified QoS properties, (2) Correctly managed related & interacting QoS, and (2) Correctly managed related & interacting QoS, and (3) Implementation artifacts that accurately represent design 15

  12. DQML Related Publications & Presentations Book Chapter Poster Publications 1. Hoffert, J., Schmidt, D., & Gokhale, A. (2011). 4. Hoffert, J., Dabholkar, A., Gokhale, A., & Productivity Analysis for the Distributed QoS Schmidt, D. (2007, March). Enhancing Modeling Language. Model-Driven Domain Security in Ultra-Large Scale (ULS) Analysis & Software Development: Systems using Domain-specific Modeling. Spring 2007 Conference for Team for Architectures & Functions. Ed. Dr. Janis Osis & Dr. Erika Asnina, Riga Technical University, Research in Ubiquitous Secure Latvia. Technology (TRUST), Berkeley, CA. 5. Hoffert, J., Schmidt, D., Balakrishnan, M., Conference Publications & Birman, K. (2008, April). Trustworthy 2. Hoffert, J., Schmidt, D., & Gokhale, A. (2007, Conferencing via Domain-specific June). A QoS Policy Configuration Modeling Modeling & Low Latency Reliable Language for Publish/Subscribe Middleware Protocols. Spring 2008 Conference for Platforms. Proceedings of the Inaugural Team for Research in Ubiquitous Secure International Conference on Distributed Event- Technology (TRUST), Berkeley, CA. Based Systems (DEBS) , Toronto, Canada. 6. Hoffert, J., Gokhale, A. & Schmidt, D. 3. Hoffert, J., Schmidt, D., & Gokhale, A. (2008, (2007, September). QoS Management in November). DQML: A Modeling Language for Publish/Subscribe Systems using Domain- Configuring Distributed Publish/Subscribe specific Modeling. ACM/IEEE 10th Quality of Service Policies. Proceedings of the International Conference on Model Driven 10th International Symposium on Distributed Engineering Languages & Systems Objects, Middleware, & Applications (DOA) , (MoDELS), Nashville, TN. Monterrey, Mexico. First Author 16

  13. FLEXible Middleware & Transports (FLEXMAT) Focus Area 1 Focus Area 2 Focus Area 3 Focus Area 4 Evaluate transport protocols with multiple operating environments using FLEXMAT testbed: • OpenSplice, OpenDDS MuxModule • Various # of senders, % loss, sending rate MulticastModule FLEXMAT … Testbed • Standard & custom protocols SequencerModule • Leverage FLEXMAT testbed ANT Framework integrated with DDS implementations • Leverage composite QoS metrics FLEXMAT addresses the challenges of (1) Supporting multiple “antagonistic” QoS via new, custom protocols & (2) Understanding how environments affect multiple QoS concerns 18

Recommend


More recommend