Software Lifecycle Software Lifecycle Basics Lifecycle Models Methods and Tools.
Lifecycle Management of Service-based Applications on Multi-Clouds: A Research Roadmap
-
Upload
george-baryannis -
Category
Travel
-
view
707 -
download
0
description
Transcript of Lifecycle Management of Service-based Applications on Multi-Clouds: A Research Roadmap
Lifecycle Management of Service-based Applications on Multi-Clouds: A Research RoadmapGeorge Baryannis, Panagiotis Garefalakis, Kyriakos Kritikos, Kostas Magoutis, Antonis Papaioannou, Dimitris Plexousakis and Chrysostomos Zeginis
Institute of Computer ScienceFoundation for Research and Technology - Hellas
International Workshop on Multi-Cloud Applications and Federated Clouds (MultiCloud 2013), April 22nd 2013, Prague, Czech Republic
Outline
• Introduction– Multi-Cloud Deployment and Lifecycle– Motivating Example
• Application Lifecycle– Requirements Specification– Infrastructure Description– Matchmaking– Monitoring and Adaptation
• ConclusionsMultiCloud 2013 April 22nd 2013, Prague, Czech Republic 2
Research Challenges
Related Work
Proposed Solution
Cloud Application Deployment
• Each SaaS deployed on a different Cloud provider
• Heterogeneity at a coarse level of granularity
SaaS Composition
• All Cloud resources tied to a single Cloud provider
• Preferable for tightly-coupled resources
Single-Cloud
• Deployment on multiple, heterogeneous providers
• Different requirements may be best satisfied by different providers
Multi-Cloud
MultiCloud 2013 April 22nd 2013, Prague, Czech Republic 3
Motivating Example
MultiCloud 2013 April 22nd 2013, Prague, Czech Republic 4
• TA: high computation and availability, moderate storage throughput
• TM: storage capacity, close to TD
• TD: no storage, close to TM
Multi-Cloud Application Lifecycle
MultiCloud 2013 April 22nd 2013, Prague, Czech Republic 5
Related Approaches
• Cloud application lifecycle can be partly addressed by Cloud brokerage frameworks, however– there is no Multi-Cloud support– some concentrate on specific application types
• TOSCA addresses the complete lifecycle of an application, employing topologies and orchestrations to model it and process models (plans) to manage it– Plans refer to the deployment and termination of the application as
well as its modification• The benefits of TOSCA can be transferred to the Multi-Cloud
context by unifying it with Cloud description languages– Combining the generality of addressing multiple Cloud domains with
lifecycle management based on TOSCA plansMultiCloud 2013 April 22nd 2013, Prague, Czech Republic 6
Requirements Specification (1)
• Resource requirements need to be clearly expressed, including capturing dependencies between components or tasks
• Current approaches such as PIM4Cloud DSL address basic deployment requirements, without focusing on formality and detail, or resource dependencies
MultiCloud 2013 April 22nd 2013, Prague, Czech Republic 7
E. Brandtzaeg, M. Parastoo, and S. Mosser. Towards a Domain-Specific Language to Deploy Applications in the Clouds. In Third International Conference on Cloud Computing, GRIDs, and Virtualization(CLOUD COMPUTING) , pages 213–218, Nice, 2012
Requirements Specification (2)
• We propose to enhance existing models in two directions– Explicit modeling of component behavior anddependencies
MultiCloud 2013 April 22nd 2013, Prague, Czech Republic 8
– Use predicate constraints to express deployment requirements
– “high”, “medium”, “low”, imply a cross-platform categorization• alternatively one can
express exact values (e.g. in relation to a particular benchmark)
Infrastructure Description (1)
• Heterogeneity in Multi-Clouds requires infrastructure descriptions to be uniform and platform-independent– Going beyond simple properties– Allowing for comparisons across Cloud providers– Facilitating automated matchmaking
• Current efforts do not address these challenges– PIM4Cloud offers too low-level descriptions (textual
descriptions of CPU frequency/cores and memory size)– Amazon’s CloudFormation is platform-specific– Cloudify offers some lifecycle support but does not provide
automated matchmakingMultiCloud 2013 April 22nd 2013, Prague, Czech Republic 9
Infrastructure Description (2)
• We aim for cross-platform comparisons of Cloud resources, by employing benchmarking
• A vector-based performance profile is created for each offered VM, containing benchmark results focusing on four dimensions:– CPU, disk and memory I/O performance– overall performance, based on benchmarking
systems such as SPECjvm2008 and Unixbench
MultiCloud 2013 April 22nd 2013, Prague, Czech Republic 10
Infrastructure Description (3)
• Clustering techniques are used to allow a cross-platform categorization of resources to different classes of service (e.g. small, medium, large)
– A cost-normalized view can be used alternatively• To facilitate automated matchmaking, we align
requirement and infrastructure specifications by employing logic-based formal specifications for both
MultiCloud 2013 April 22nd 2013, Prague, Czech Republic 11
Matchmaking (1)
• Current approaches for matchmaking between deployment requirement and Cloud offerings face several challenges– May be completely manual and thus inefficient– If automation is attempted, it is only achieved by
oversimplification, expressing requirements at the level of resource descriptions
– No support for Multi-Cloud provisioning
MultiCloud 2013 April 22nd 2013, Prague, Czech Republic 12
Matchmaking (2)
• We propose the use of constraint satisfaction rules to drive matchmaking– Can be high-level, or
application/domain-specific– Expressed by deployment experts
or derived from learning processes [2] based on deployment history
MultiCloud 2013 April 22nd 2013, Prague, Czech Republic 13
• Constraint logic programming techniques can then be employed to realize matchmaking– solving an optimization problem, providing the best possible
deployment solution satisfying the constraints
Matchmaking (3)
• The matchmaking process requires the use of the previously described logic-based formal specifications for requirements and infrastructure
• Deployment plans are ranked based on the importance of the optimization criteria (e.g. total execution time, availability)
• Rule-based reasoning in this case is expected to be efficient:– limited number of dimensions– absence of looping rules– low probability of contradiction
MultiCloud 2013 April 22nd 2013, Prague, Czech Republic 14
Monitoring and Adaptation (1)
• Monitoring and adaptation in the Multi-Cloud context has yet to address the following challenges:– Uniform, cross-platform support among different
Cloud providers– Cross-layer (IaaS, PaaS and SaaS) monitoring and
alignment of monitored events, – Cross-layer coordination of adaptation actions– Use of both proactive and reactive adaptation
policiesMultiCloud 2013 April 22nd 2013, Prague, Czech Republic 15
Monitoring and Adaptation (2)
• We plan to extend previous work on cross-layer event-based monitoring and adaptation of SBAs [3] to the Multi-Cloud context– Metric derivation trees (MDTs) allow for detecting which low-
level metrics cause violations detected by monitoring– Metric matching [1] algorithms can be used to correlate between
different but related terms used in various Cloud settings– Event-to-action correlation rules map (possibly) cross-layer event
patterns to adaptation actions (re-deployment):
– Adaptation action history can be used to re-design the application (or modify requirements) when excessive repetition is detected
MultiCloud 2013 April 22nd 2013, Prague, Czech Republic 16
Conclusions
• A research roadmap for addressing current challenges in the lifecycle management of SBAs on Multi-Clouds– Capture dependencies and behavioral attributes– Logic-based approach, expressing requirements
and capabilities as a set of predicate constraints– Matchmaking via a constraint logic programming
approach– Realize cross-layer and cross-platform monitoring
and adaptationMultiCloud 2013 April 22nd 2013, Prague, Czech Republic 17
Questions
MultiCloud 2013 April 22nd 2013, Prague, Czech Republic 18
References
1. K. Kritikos and D. Plexousakis. Semantic QoS Metric Matching. In ECOWS, pages 265–274, Zurich, Switzerland, 2006. IEEE Computer Society.
2. K. Magoutis, M. Devarakonda, N. Joukov, and N. Vogl. Galapagos: Model-Driven Discovery of End-to-End Application-Storage Relationships in Distributed Systems. IBM Systems Journal, 52(4/5):367–378, 2008.
3. C. Zeginis, K. Konsolaki, K. Kritikos, and D. Plexousakis. Towards Proactive Cross-Layer Service Adaptation. In WISE, pages 704–711, 2012.
MultiCloud 2013 April 22nd 2013, Prague, Czech Republic 19