smart cloud federation simulations with cloudsim
play

Smart Cloud Federation Simulations with CloudSim Gaetano F. - PowerPoint PPT Presentation

Smart Cloud Federation Simulations with CloudSim Gaetano F. Anastasi, Emanuele Carlini, Patrizio Dazzi Contrail is co-funded by the EC 7th Framework Programme under Grant agreement nr. 257438 National Research Council of Italy Clouds are


  1. Smart Cloud Federation Simulations with CloudSim Gaetano F. Anastasi, Emanuele Carlini, Patrizio Dazzi Contrail is co-funded by the EC 7th Framework Programme under Grant agreement nr. 257438 National Research Council of Italy

  2. Clouds are spreading... • Several different Cloud Providers and Technologies - providing very different kind of resources both from the point of view of: Abstraction ✤ Features ✤ Usage model ✤ • Adoption in different sectors • Towards a real market of Clouds National Research Council of Italy

  3. ...but are current Clouds issue-free ? • Some Cloud Provides could provide only specialized resources • Who is cheaper for some resources could be expensive for others • Cloud Providers could be located only in one (or just a few) country • User lock-in issue in Clouds National Research Council of Italy

  4. How to solve ? • A Need for Super-entities seamlessly managing several Cloud Providers • Federation of Clouds is a potential solution - The Contrail approach to Cloud Federations - Resource Allocation in Cloud Federations is important - Testing it could be not very simple National Research Council of Italy

  5. Simulate to Test Solutions • Estensively test different allocation solutions in “real” Cloud Federation is complex and expensive - Several possible environments / technologies - Different kinds of applications - $$$ it costs money $$$ • Simulators are part of the solution National Research Council of Italy

  6. What do we want to simulate in a Federation ? • Different Kinds of - Applications inter-arrival ✤ constraints ✤ - Clouds Resources ✤ Cost Models ✤ • Different Strategies for - Monitoring - Allocation National Research Council of Italy

  7. What did we do ? • A model for describing Cloud Federations • An extension to CloudSim addressing Contrail like Cloud Federations • Validated our approach conducting some tests National Research Council of Italy

  8. The Cloud Federation Model National Research Council of Italy

  9. How we model Cloud Federations • Application Model • Resource Model • Application Queue Model • Resource Monitoring Model • Resource Allocation Model National Research Council of Italy

  10. Application Model • Graph-based Representation CPU = ... RAM = ... - Node = Appliance SLA terms = ... N/W Bandwidth = ... - Edge = Information Exchange • Functional Requirements • SLA Terms • Network Requirements National Research Council of Italy

  11. Resource Model • Organized in Datacenters • IaaS resources - Presented as VM slots • Internet Estimator for modeling interconnections National Research Council of Italy

  12. Application Queue Model • Models the arrival of Applications - Inter-arrival time - Information provided ✤ e.g. amount, quality, ... National Research Council of Italy

  13. Resource Monitoring Model • Amount and Quality of information about resources availability - Update frequency - Aggregation style National Research Council of Italy

  14. Resource Allocation Model • Association of Resources to Applications trix - Appliance by Appliance   c 1 , 1 c 1 ,P · · · - . . ... Allocation Matrix . . Y =   . .   - c i,j means: c N, 1 c N,P · · · where the element c re j-th datacenter able to run ✤ the i-th appliance • Computed using both static and dynamic information National Research Council of Italy

  15. The CloudSim Extension National Research Council of Italy

  16. SmartFed Simulator • Built on top of CloudSim exploiting its: - Datacenter - Broker - CloudLet • CloudSim proven to be well-conceived for reusing and flexibility • Why we had to add something ? National Research Council of Italy

  17. What is missing in CloudSim ? • Datacenters characterization with non-functional features, - e.g. reputation estimation • Brokers with complex allocation strategies - integrated with the whole SmartFed eco-system • CloudLets characterized by explicit relationships with other CloudLets National Research Council of Italy

  18. SmartFed Eco-System SmartFed • Application Queue Monitoring SLA Storage <<use>> • Application Queue Application <<use>> <<use>> <<use>> <<use>> <<use>> <<use>> <<use>> Federation • Internet Estimator Application Application Allocator Datacenter Vertex Edge Reputation allocate • Monitoring Hub <<use>> <<use>> <<use>> Broker Datacenter CloudLet • Allocator CloudSim National Research Council of Italy

  19. The Validation of the Approach National Research Council of Italy

  20. Hands on SmartFed • Validation of SmartFed - A sample, very classic application - Different sizes of applications and datacenters - Very simple allocation algorithm National Research Council of Italy

  21. A sample Application • Three-tier applications CPU type x64 CPU Freq. 2 Ghz Cores 2 RAM 2 GB • Web front-end Web Frontend (Presentation tier) Latency 1000 ms • Business logic Bandwidth 1024 Kb/s CPU type x64 Security SSL CPU Freq. 3 Ghz Cores 4 RAM 16 GB ApplicationServer • Database (Business logic tier) Latency 1000 ms Bandwidth 512 Kb/s CPU type x64 Security SSL • 10, 100 and 1000 CPU Freq. 2 Ghz Cores 2 Database RAM 4 GB appliances per application (Persistency tier) National Research Council of Italy

  22. Machines distribution into Clouds (Datacenter entities) • Two different kind of distributions - Uniform: each datacenter owns the same ✤ number of machines each datacenter has enough ✤ machines to run an instance of the application - Normal: the # of machines in each datacenter follows a normal distribution ✤ some datacenters are not be able to host the application ✤ National Research Council of Italy

  23. Allocation Algorithm • When the Application is received: - is decomposed in its VMs - a Datacenter d is chosen randomly among the ones able to run one or more VMs - If d has enough room is chosen - When d is full, another datacenter is chosen - When all the VMs have been assigned, it terminates National Research Council of Italy

  24. Experimental Results                           10 Datacenters 100 Datacenters National Research Council of Italy

  25. Conclusions • Modeling of application and cloud resources targeting (Federation of) Cloud platforms - Contrail in particular but a pretty general approach • Enhriched CloudSim to support - more complex SLA model - mechanisms to support off-line mapping of applications • Validation of the solution National Research Council of Italy

  26. Future Works • Extensive validation - testing different allocation strategies - validation against different Cloud Simulators e.g. iCanCloud ✤ • Exploiting different kind of applications and resources - possibly using information about real ones • Integration of Security Aspects • Enhanced Modeling of Queues - e.g. with Markov models National Research Council of Italy

  27. Acknowledgements contrail is co-funded by the EC 7th Framework Programme Funded under: FP7 (Seventh Framework Programme) Area: Internet of Services, Software & Virtualization http://contrail-project.eu (ICT -2009.1.2) Project reference: 257438 Total cost: 11,29 million euro EU contribution: 8,3 million euro Execution: From 2010-10-01 untill 2013-09-30 Duration: 36 months Contract type: Collaborative project (generic) National Research Council of Italy

  28. Questions ? National Research Council of Italy

Recommend


More recommend