Dana Petcu (IeAT), Nicolas Ferry (SINTEF), Marco Miglierina (POLIMI), Alexander Gunka (BOC), Florin Picioroaga (SIEMENS), Marcos Almeida (Softeam)

Transcription

1 Grant Agreement N FP Title: Authors: Editor: Reviewers: Identifier: Nature: Report on best practices Initial version Dana Petcu (IeAT), Nicolas Ferry (SINTEF), Marco Miglierina (POLIMI), Alexander Gunka (BOC), Florin Picioroaga (SIEMENS), Marcos Almeida (Softeam) Dana Petcu (IeAT) Oliver Barreto (ATOS), Antonin Abherve (Softeam) Deliverable D Report Version: 1 Date: 31 March 2014 Status: Diss. level: Final Public Executive Summary This deliverable presents the best practices that are to be followed to maximize the usefulness of the MODAClouds solution. The report is the result of a special activity of task Communications of the project, activity, which serves as a wrap-up of the knowledge gained throughout the project lifetime. The report intends to spread the lessons learned as an introductory guide for MODAClouds future users, partners or parties interested to exploit the results. The report includes a study of the current state of the art, the results of the MODAClouds innovative activities at the middle of the project, and a comprehensive repository of best practices derived from the project activities. Copyright 2014 by the MODAClouds consortium All rights reserved. The research leading to these results has received funding from the European Community's Seventh Framework Programme [FP7/ ] under grant agreement n (MODAClouds).

2 Members of the MODAClouds consortium: Politecnico di Milano Stiftelsen Sintef Institutul E-Austria Timisoara Imperial College of Science, Technology and Medicine SOFTEAM Siemens Program and System Engineering BOC Information Systems GMBH Flexiant Limited ATOS Spain S.A. CA Technologies Development Spain S.A. Italy Norway Romania United Kingdom France Romania Austria United Kingdom Spain Spain Published MODAClouds documents These documents are all available from the project website located at Public Final version 1.0, 31/03/2014 2

3 Contents 1 INTRODUCTION CONTEXT AND OBJECTIVE WHAT WE EXPECT FROM A BEST PRACTICE REPORT STRUCTURE OF THE DOCUMENT THE CONTEXT THE RAISE OF MULTI- CLOUDS SOFTWARE SUPPORT FOR MULTI- CLOUDS REQUIREMENTS AND CHALLENGES INTRODUCTORY GUIDE TO MODACLOUDS THE SCOPE MODACLOUDS' CONCEPTS IMPLEMENTATION STATUS PROOF- OF- CONCEPT APPLICATIONS BEST PRACTICE REPOSITORY PRELIMINARIES HOT TOPICS Standard adoption Interoperability and portability Avoid lock- in TRANSFORM Roadmap Consulting Scenarios PRACTICE Open Cloud Ecosystem Cloud Migration Management Cloud Service Brokerage Cloud Management and Analytics SOLUTIONS Software Services CONCLUSIONS REFERENCES Public Final Version 1.0, 31/03/2014 3

4 1 Introduction 1.1 Context and objective This deliverable is the result of a special activity of Task 11.3 (communications) of the project, the activity which serves as a wrap-up of the knowledge gained throughout the project lifetime. With a study of the current state of the art, the conclusions and results of the MODACLOUDS innovations, the best practices report intends to spread the lessons learned as an introductory guide for MODACLOUDS future users, partners or parties interested to exploit the results. 1.2 What we expect from a best practice report There is today a concern in any country to understand how to become more competitive in science and technology. Because of many factors interrelate and outcomes are context specific, no single optimal solution is at hand to merely plug in to fit a particular need. This lack of universally applicable solutions and lessons makes it even more important to study and learn from the experience of others, so as to be able to translate lessons in such a way as to let them inspire constructive reform. Best practice is a very popular term within the IT community especially as marketing concept. Best practices are used to maintain quality through reported experiences and can be based on selfassessment or benchmarking. In our case will be self-assessed. In what follows we consider that a best practice: is a method or technique that has consistently shown results superior to those achieved with other means, and that is expected to be used as a benchmark; balances the need for very careful description in academic literature with the desire for maximum clarity by the practitioners; defines a number of practical rules and examples of successful implementations where these rules are applied (rules of thumb) and which are recommended to be followed. There are many examples where a practice that is considered best in one context is contested within another. The followers should understand the context and then apply the best practice within that context we refer therefore to a contextual best practice. More exactly, the best practice is bound to a set of conditions and circumstances to achieve the result. Moreover, a best practice in IT can be obsolete over time due to the technological advancements. 1.3 Structure of the document This document is structured as follows. The next section is describing the context in which MODAClouds evolves and which is motivating it. Section 3 is presenting the state-of-the-art in what concerns the MODAClouds developments. Section 4 is exposing the proposed best practice repository. Finally, Section 5 includes few conclusions. Public Final version 1.0, 31/03/2014 4

5 2 The context 2.1 The raise of Multi-Clouds The usage of services and resources from multiple Clouds is push forward by the needs of their consumers or their providers. The main expectation of the Cloud service providers that one size fits all needs is not always matching the service consumer requirements. Several reasons are pushing the multiple Clouds usage. Most of the ones mentioned recently mentioned are displayed in Figure 1. Consume different services for their particularities not provided elsewhere Ensure backup-ups to deal with disasters or scheduled inactivity Enhance own Cloud resource and service offers, based on agreements with other providers Avoid the dependence on only one external provider Act as intermediary Replicate applications/services consuming services from different Clouds to ensure their high availability Simultaneous usage of services from multiple Clouds Driving forces for multiple Clouds Serial usage of services from multiple Clouds Follow the constraints, like new locations or laws React to changes of the offers by the providers Deal with the peaks in service and resource requests using external ones, on demand basis Optimize costs or improve quality of services Figure 1. Reasons for using multiple Clouds Different actors can have different reasons for using the multiple Clouds. Table 1 proposes a matrix of potential interest for four actors: service providers, service brokers (third party), service or application developers, service or application consumers. Table 1. Interests split between multiple actors in using multiple Clouds Service provider Service broker Service developer Service consumer Particularities Enhance offer Intermediary Availability Peaks Backup Non-dependence Constraints Changes Costs & QoS Despite the fact that it practically emerged with the Cloud computing concept, the field of multiple Clouds is nowadays still in an infancy stage, mainly due to the diversity of the approaches of the Public Final Version 1.0, 31/03/2014 5

6 concept implementation. Figure 2 points to the main categories of multiple Clouds, while Table 2 is including their definitions. A deeper classification is discussed in [1]. Library-based Multi-Clouds Distributed Clouds Service-based Multi-Clouds Dynamic Federation Hybrid Clouds Multi-Clouds Multiple Clouds Cloud Federations Vertical Federations Clouds of Clouds Multi-tier Federations Figure 2. Multiple Clouds: main categories Table 2. Main categories of multiple Clouds Category Cloud Federation Multi-Cloud Distributed Cloud Dynamic Federation Vertical Federation Multi-tier Cloud Library-based Multi-Cloud Service-based Multi-Cloud Hybrid Cloud Clouds-of-Clouds Description Multiple Cloud delivery model assuming a formal agreement between the Cloud providers sub-contracting capacity from other providers or offering spare capacity to the federated group, in a non-transparent mode for their service consumers Multiple Cloud delivery model assuming no priori agreement between the Cloud providers and a third party responsible for provider contacting, consumption negotiation, SLA monitoring, inter-provider networking, code and data migration Horizontal Federation based on a wide area network such as the Internet Federation in which the providers collaborate dynamically to gain economies of scale and enlargements of their capabilities, and the inter-connections are dynamic in terms of joins and leaves, autonomous in terms of access control policy for each Cloud, and distributed in terms of trust establishment mechanisms like authorization delegation or role mapping across domains Federation of Cloud providers offering services to different layers, IaaS, PaaS and SaaS Two or more Clouds tightly coupled with a third Cloud that has advanced control over remote services, full access to monitoring information and advanced networking features Multi-Cloud based on a library that offers an uniform interface to the Clouds and a broker that takes care of provisioning of services across Clouds Multi-Cloud based on a service hosted externally or in-house by the customers and including a broker using service level agreement or a set of provisioning rules Set of services combined from a Private Cloud and one or more Public Clouds Set of Cloud services that are composed of one or more services from other Clouds We are focusing in MODAClouds on Multi-Clouds. The reason is the advantage of the Multi-Cloud versus the Cloud Federation: it assumes that there is no prior agreement between the Cloud providers. A third party (even the consumer) is responsible for the services. This third party contacts the service providers, negotiates the terms of service consumption, monitors the fulfillment of the service level agreements, and triggers the migration of codes, data and networking from one provider to another. Public Final version 1.0, 31/03/2014 6

7 2.2 Software support for Multi-Clouds We enumerate in what follows the most prominent approaches for supporting Multi-Clouds. For a detailed description, please follow [1]. The most known library-based approaches for Multi-Clouds support are jclouds, libcloud, δ-cloud. Jclouds 1 is an open source Java library designed to support the portability of Java applications, which allows the uniform access to the resources from various IaaS providers. Libcloud 2 is a Python library that abstract the differences among the programming interfaces of Cloud services. δ -cloud 3 is a REST-based API written in Ruby which allows also the connections to various Cloud resources. Service-based approach for Multi-Cloud can also be classified in two categories: hosted or deployable. The most known hosted services for Multi-Cloud are the commercial offers of RighScale, Kavoo and Enstratius. RightScale 4 is offering a management platform for the control and administration of deployments in multiple Clouds; its Multi-Cloud Engine is able to broker capabilities related to virtual machine placement in Public Clouds. Kaavo 5 allows the management of distributed applications and workloads in various Clouds. Enstratius 6, allows the management, automation and governance of resource consumption based on the services from various Cloud providers. Several deployable services are results of open-source projects like Aoleus, mosaic, Cloud4SOA or OPTIMIS (last three are FP7-ICT projects). Aeolus 7 is an open-source cloud management software written in Ruby and provided for Linux systems by RedHat and it is based on the δ-cloud library. mosaic 8 offers an open-source API aining to ensure portability of applications consuming infrastructure services and a deployable PaaS allowing the deployment and the life-cycle control of applications consuming infrastructure services. Cloud4SOA 9 has deal with portability of applications between PaaSs by relying upon semantic technologies. OPTIMIS Toolkit 10 offers a deployable PaaS that allows Cloud service provisioning and the management of the life-cycle of the services. Other current FP7-ICT projects related to multiple Clouds are the followings. 4CaaST 11 has building a Cloud blueprint for marketplaces. REMICS 12 has demonstrated how to migrate legacy applications (desktop of standard client-server) to Cloud based architecture using model-driven engineering. Vision Cloud 13 has deal with vendor-agnostic data management services. TClouds 14 has investigated the trust management in Clouds-of-Clouds. Currently Broker@Cloud 15 develops methods and mechanisms for quality assurance and optimization of software-based services, while CloudSpaces 16 is looking into interoperability mechanisms between Personal Clouds. MODAClouds is trying to follow the model-driven Blueprint approach and is relying upon the solutions of mosaic and Cloud4SOA for Multi-Clouds. Beyond MODAClouds and REMICS, model-driven engineering is proposed to be used in multiple Clouds also by ARTIST 17 and PaaSage libcloud.apache.org 3 deltacloud.apache.org enstratius.com 7 aeolusproject.org 8 bitbucket.org/mosaic 9 github.com/cloud4soa/cloud4soa caast.morfeo-project.org Public Final Version 1.0, 31/03/2014 7

8 The Cloud brokers are playing an important role in Multi-Clouds. The most known brokers are: SplotCloud, Scalr and Stratos. SpotCloud 19 provides a marketplace for infrastructure service and a matching service with the client requirements. Scalr 20 provides deployment of virtual machines in various Clouds and includes automated triggers to scale up and down. Stratos 21 offers single sign-on and monitors resource consumption and the fulfillment of service level agreements and offers autoscaling mechanisms. 2.3 Requirements and challenges Table 3 enumerates the main requirements for a Multi-Cloud support system. Their motivation and more details are provided in [2]. Table 3. Requirements for a Multi-Clouds software support system Stage of application Development support Deployment support Execution support Requirements Resource/service (meta-)management (portal, service, interface) Services that are Cloud vendor agnostic An interface for describing functional and non-functional requirements An Integration-as-a-Service or service aggregators to combine services from different Clouds Selection service for the Cloud services to be used A service and resource meta-allocator A (meta-) scheduler, a (meta-)load-balancing or auto-scaling mechanisms Authentication services for single sign-on or credentials repositories Automated procedures for deployments Deployer of components of applications in multiple Clouds Deployer on Private Clouds to enable testing, debugging, or privacy Network overlay mechanisms to overcome limited connectivity A search engine based on a taxonomy or using semantic processing A match-making or brokering service A service and resource selection interface A recommendation system, a trust management system or a reputation management system A (meta-)monitoring service for the deployed applications and Cloud resource consumptions Control of the full life-cycle of the deployed applications Metering of the degree of fulfillment of the service level agreements Despite the intensive work in the field of Multi-Clouds, as reflected also in the availability of the above described tools and services, the main actors requirements are only partially fulfilled by them. In order to make the usage of services from Multi-Clouds a reality, several technical barriers should be over-passed, like interoperability and portability, data and services mobility or middleware openness spotcloud.com 20 scalr.net 21 wso2.com/cloud/stratos/ Public Final version 1.0, 31/03/2014 8

9 Moreover, the desired semi-automated guidance through the variety of the offers, based on monitoring tools for the quality of services, is not yet technically possible. Furthermore, the differences between the current APIs are hindering the easy composition or configuration of service to be consumed from Multi-Clouds. Adapters and abstraction libraries which provides the ability for a new Cloud service provider to easily connect to a certain framework are needed. This request is not coming from the primary market, organizations looking for Multi-Clouds, but from a secondary market, enabling cloud providers to be eligible to enter in an already established marketplace. Another issue that is directly treated by MODAClouds is the need of simplification of the Cloud application development, deployment, operation and management and is related to the DevOps concept that has emerged recently in conjunction with Cloud computing. Public Final Version 1.0, 31/03/2014 9

10 3 Introductory guide to MODAClouds 3.1 The scope The quality of cloud services is currently a widely debated subject. Its modeling at application design time and its evaluation at the application run-time are currently subjects for intensive research and developments. Model-driven engineering techniques can help in finding a proper solution. We discuss in this chapter the early results of an initiative for providing an open source, integrated and modeldriven development environment for the high-level design, early prototyping, semi-automatic code generation, and automatic deployment of applications on multiple clouds, with guaranteed quality of service. Cloud business models and technologies are in their initial hype and characterized by critical early stage issues, which pose specific challenges and require advanced software engineering methods. Model-driven development combined with model-driven risk analysis and quality prediction can theoretically enable application developers to specify models of Cloud services that are vendor agnostic and which are enriched with quality parameters, implement these, perform quality prediction, monitor applications at run-time and optimize them based on the feedback, thus filling the gap between design and run-time. To prove that this idea can work in practice is the main aim of a recently started research initiative, namely MODAClouds. Model-Driven Approach for design and execution of applications on multiple Clouds (shortly MODAClouds) aims to provide methods, a decision support system, an open source IDE (Integrated Development Environment) and run-time environment for the high-level design, early prototyping, semi-automatic code generation, and automatic deployment of applications on multi-clouds with guaranteed QoS (Quality of Service). MODAClouds concept was presented for the first time in [3] paper which included a motivating scenario. Several more recent papers, mentioned in what follows, provide details about particular solutions and components. A deep analysis of the state of the art was done in the frame of MODAClouds' public deliverables 22 as follows: D2.2 relative to the modeling of costs and benefits, D4.1 relative to the Cloud service modelling, D5.1 relative to the QoS and performance modelling at design time, D6.1 relative to the Cloud monitoring techniques, QoS management at run-time and automatic deployment in Cloud environments. We mention here only the most significant and closest initiatives. The Service Level Agreement (SLA) is widely used today for managing QoS in Cloud environments. A QoS attribute in a SLA describes a specific measurable aspect of the quality of service. QoS management steps are discussed e.g. in [4]. The technologies of interest for SLAs and QoS specifications in MODAClouds are SLAang 23 and QML [5]. The latest is an extension of the UML which enables multi-category QoS specification for components in distributed object systems and addresses specific QoS metrics such as reliability, performance, security, availability (not adapted to Clouds). SLA@SOI 24 framework is based on a SLA-enabling reference architecture suitable for both new and existing service-oriented systems and Cloud infrastructures. It consists of a suite of open-source software and components to implement SLA-aware solutions [6]. It covers most of the levels of the service provisioning process; uclslang.sourceforge.net 24 Public Final version 1.0, 31/03/

11 however it is not standards-compliant and defines a proprietary SLA stack [7]. Later on, Cloud4SOA adapted the SLA@SOI conceptual framework for PaaS environments. Similarly, in mosaic, SLA@SOI concept was used at IaaS level [8]. SLAs in Contrail 25 project are used to specify QoS for performance, availability and the quality of protection (QoP) for security guarantees for infrastructure components; the starting point is again the SLA@SOI conceptual framework, and SLA template creation, browsing and querying, SLA execution planning and adjustment were implemented [9]. PaaSage project is providing novel techniques to evaluate the distributed application deployments in Multi-Clouds [10]. The most recent project, SPECS 26 aims to enhance the security definition is SLAs, to monitor the SLA compliance and to ensure its enforcement. Despite the fact that all of these initiatives are exposing incipient forms of monitoring systems and feedback mechanisms, none of them has a comprehensive technical solution for QoS management as MODAClouds currently provides. 3.2 MODAClouds concepts In order to ensure service vendor agnosticism in a market in which diversity of the service interfaces is hindering the adoption by customers, a certain level of abstraction is needed at the application design time. In MODAClouds, Cloud services are represented and handled in a specific modelling language, namely MODACloudML [11]. This is an extension of CloudML 27, which provides a domain-specific modelling language along with a run-time environment that facilitate the specification of provisioning, deployment, and adaptation concerns of multi-cloud systems at design-time and their enactment at run-time. The models manipulated by the MODACloudML are grouped within the dashed box of Figure 3. More details can be found in [12,13]. Figure 3. Models in MODACloudML The MODACloudML architecture is inspired by the OMG Model-Driven Architecture 28 (MDA), a model-based approach for the development of software systems. The specific MDA relies on three types of models for three layers of abstractions. These layers, from the more abstract to the more detailed, are CCIM, CPIM and CPSM: Cloud Computational Independent Model (CCIM) describes what the system is expected to, hiding all the technical details related to the implementation of the system. 25 contrail-project.eu/sla 26 specs-project.eu Public Final Version 1.0, 31/03/

12 Cloud Platform Independent Model (CPIM) describes views of the systems in a platform independent manner so that it can be mapped to several platforms at the CPSM levels. Cloud Platform Specific Model (CPSM) refines the CPIM with technical details required for specifying how the system can use a specific platform. MODAClouds software stack is depicted in Figure 4. The most recent overview paper explaining in details the MODAClouds architecture is [14]. Figure 4. MODAClouds Software Stack MODACloudML is supported by the MODAClouds IDE, which provides the functional, operational and data modelling environments, as well as some modules enabling the analysis of non-functional Public Final version 1.0, 31/03/

13 characteristics of a multi-cloud. From the point of view of the user, the IDE is the main piece of software to be used at design time. It realizes the model-driven engineering approach proposed by the MODAClouds, and its main output is the set of models and artefacts required by the runtime components to deploy, monitor and adapt cloud-based applications. The IDE includes also decision support tools to guide the construction of the application's models. Furthermore, it provides a common access point for a set of tools (its core functionality), and a set of accessory functionalities to aid the use of the other tools. The IDE is composed of five tool sets: 1. The QoS Modelling and Analysis Tool set allows a quality of service engineer to model the QoS requirements of the application, to assess the performance of a given deployment and to identify the deployment configuration that minimizes costs. The results of the simulation can be used by the application developer to improve the design of the code and its data and by the application provider to improve the deployment. 2. The Functional modelling Tool set implements the MODACloudML meta-model and provides a user interface allowing the editing and storing of a MODACloudML model. It provides transformation, reverse engineering, traceability and document and code generation capabilities. 3. The Decision Making Toolkit allows a feasibility study engineer to model the costs and risks of an application architecture. His or her decisions will then impact on the deployment, QoS and design of the application. 4. The Deployment and Provisioning Component interfaces with the runtime components and allows the application provider to deploy the application. 5. The Data Mapping Component allows the application developer to design the data that is manipulated and stored by the application. Data structures can be analyzed from the point of view of the best runtime performance. 3.3 Implementation status The first two tool sets are currently available for testing (see bellow), while the next two are still in test phase (the decision tool is presented in [15]), while the last one is under development (a proof of concept is described in the deliverable D4.4.1). The QoS Modelling & Analysis Tool is composed from LINE and SPACE4Cloud. LINE 29 is a tool for the performance analysis of cloud applications. It has been designed to automatically build and solve layered queuing network performance models, and is able to provide accurate estimates of relevant performance measures such as application response time or server utilisation. It can therefore be used at design time to diagnose whether the deployment characteristics are adequate to attain the desired QoS levels. LINE stands apart from other tools available for performance modelling for a number of reasons: (1) in addition to provide average performance measures, it has been designed to compute response time distributions, which can be directly used to assess percentile service level agreements; (2) features describing the specificities of cloud applications (e.g. random environments modeling reliability and multi-tenancy, or general service times representing the resource demands posed by the very broad range of cloud applications); (3) able to compute transient performance measures, capturing the effect of temporary conditions, such as workload spikes or failures of the application components. 29 code.google.com/p/line Public Final Version 1.0, 31/03/

14 SPACE4Cloud 30 (System PerformAnce and Cost Evaluation on Cloud) [16] is a tool for the specification, assessment and optimisation of QoS characteristic of cloud applications. It allows users to describe the architecture of their application by means of models both at CCIM and CPIM level following the model-driven paradigm. These models are then evaluated against a user-defined workload in order to assess both performance and cost of the modelled solution. The tool is built on top of the Palladio Bench modelling environment 31, enriching its modelling capabilities, allowing more expressiveness in the definition of the resource environment and the specification of the workload. Moreover, it implements state-of-the-art metaheuristic techniques to effectively and efficiently explore the space of possible alternative configurations. For each configuration involved in the search process the tool is also capable of evaluating the overall operative cost (it makes use of LINE for the performance evaluation). As stated earlier, the Functional modelling tool implements the MODACloudML metamodel and provides a user interface allowing the editing and storing of a MODACloudML model. It uses and enhance Modelio (as detailed in [17]). Modelio 32 is a combined UML/BPMN modeler supporting a wide range of models and diagrams. Its main features are: Java code generation, scripting language support, and extensibility (it can be extended for any language, methodology or modeling technique just by adding modules). In what concern the status of Runtime Platform development, currently first versions of the Monitoring and Execution Platforms are available, while the Self-Adaptation Platform is under development (it will be responsible for observing the monitoring data and statistics and deciding at runtime for corrective actions that can improve QoS, e.g. the deployment of new virtual machines to scale out). The main components of the Monitoring Platform are the data collectors and the data analyzers. The first ones gathers information at different levels to evaluate the QoS; the consumption of system resources (like CPU, Memory) and other QoS metrics (e.g. response time of a request) are captured. The second ones processes the information gathered by the monitoring collectors and generates highlevel statistics from them (e.g. analyze the correlation between different metrics to infer the cause of a given error in the system). They also performs statistical inference to estimate the parameters needed for the runtime QoS management models used in the Self-Adaptation Platform. Monitoring rules are received by from the IDE, define what metrics to collect and how to measure them, such as the monitoring time granularity. The platform monitors them and provides triggers and data streams to the Self-Adaptation Platform and sends feedback to the IDE. Developers and cloud system administrators can interact with the platform through graphical user interface (part of IDE; how the platform is linked with the IDE is presented in [18]) or command-line interface (a typical usage scenario is a system administrator who wants to visualize the historical performance offered by a cloud application). Data collectors are responsible for collecting monitoring data from cloud resources and applications and to associate semantic information to the data. The data is collected periodically and the monitoring period is defined by the users via configuration files. The monitoring data is encoded in RDF format and is processed the continuous query engine of C-SPARQL engine 33. Two different kinds of data analysers are featured: deterministic and statistical data analysers. The first ones process at high-speed the RDF monitoring data coming from the data collectors and try to detect onthe-fly patterns that emerge directly from the data, without the need of major transformations of the data itself. The second ones extract hidden information from the data using statistical methods and generate predictions. Details about the platform functionality are provided in [19] and the public 30 home.deib.polimi.it/gibilisco/space4cloud/space4cloud.zip forge.modelio.org/projects/modelio/files 33 streamreasoning.org/download Public Final version 1.0, 31/03/

15 deliverables related to the specifications (D6.2) and the first implementation (D6.3.1), while the source codes are available on MODAClouds repositories 34. The Execution Platform handles the deployments and execution on Cloud environments (offering IaaS and PaaS), allowing the user to gather a unified experience regardless the selected Cloud environments. Its main components are: i. platform system, build from the services which are tightly related with the MODAClouds environment; ii. infrastructure system, build from services which handle the management of the cloud service, potentially enhancing them, and providing critical services required by other components; iii. coordination system, offering services like distributed communication, coordination, data storage. In order to ensure the portability of the applications, the last two components are based on mosaic PaaS at IaaS level and Cloud4SOA at PaaS level. 3.4 Proof-of-concept applications The proof-of-concept applications are currently under development. Table 4 points to their value in validating the concepts and tool implementations. Details about the first implementation of the second use case are provided in [20]. Table 4. Use cases for proof-of-concept Case study Type Main requirements addressed to the MODAClouds' software stack Project Management Server New Application Effective resource scaling following the run-time monitoring Business Process Modelling System Legacy application Decision support system Palliative Application Care New application service Sustain a hybrid cloud with heterogeneous environments Smart City Urban Safety Planner New application service Data management, scalability and availability 34 Public Final Version 1.0, 31/03/

16 4 Best practice repository 4.1 Preliminaries Usually the Best practices are presented in clear text rather than in a tabular form. However we consider that the reuse of best case scenarios is simplified if they are organized according to a pattern or template. Therefore, in what follows we will use the best practice (BP) template proposed by [21]: [BP Template] Restrictions of applicability depending on scenarios / use cases or other circumstances and effects on other BPs Detailed action plan of features to be implemented or steps to be taken when following the BP Quantified impact of the implementation of this BP on the system or component it is being implemented into. Source material for BP, including references, experiments, etc. Actually implemented and tested configurations of the BP which performed well Extending the categories exposed to Cloud Computing Best Practices ( we consider the categories of the best practice for our repository that are presented in Figure 5. Standards adoption Open Cloud Ecosystem Interoperability and portability Avoid lock-in Hot topics Practice Cloud Migration Management Cloud Service Brokerage Roadmap Best practices categories Cloud Management and Analytics Consulting Scenarios Transform Solutions Software Services Figure 5. Best practices categories Public Final version 1.0, 31/03/

17 4.2 Hot topics Standard adoption MODAClouds is: Promoting CloudML Aligned with TOSCA Following CAMP CloudML Domain specific modeling language enhanced with a run-time environment that facilitate the specification of provisioning, deployment and adaptation at design time and enactment at runtime Used by MODAClouds IDE, Self-Adaptation Reasoner Embrace the model-driven engineering style of applications at design time Support IaaS and PaaS deployment Provide the language at runtime to ease the dynamic adaptation of cloud-based applications Strong influence on the application life-cycle: reducing the development and deployment time Tested on Amazon AWS and Flexiant FlexiScale Public Final Version 1.0, 31/03/

18 TOSCA OASIS Topology and Orchestration Specification for Cloud Applications intends to enhance the portability of cloud applications and services. It enables the interoperable description of application and infrastructure cloud services, the relationships between parts of the service, and the operational behavior of these services (e.g., deploy, patch, shutdown)--independent of the supplier creating the service, and any particular cloud provider or hosting technology CloudML and MODAClouds Execution Platform Use emerging TOSCA implementations (like OpenTOSCA, open source runtime environment for TOSCA) to ensure portability at IaaS level Compatibility with major IaaS services CAMP OASIS Cloud Application Management for Platforms advances an interoperable protocol that cloud implementers can use to package and deploy their applications. CAMP defines interfaces for self-service provisioning, monitoring, and control. Based on REST, CAMP is expected to foster an ecosystem of common tools, plugins, libraries and frameworks. CloudML and MODAClouds IDE Use emerging CAMP implementations to ensure portability at PaaS level Compatibility with major PaaS services. Public Final version 1.0, 31/03/

19 4.2.2 Interoperability and portability MODAClouds use a model-driven architecture to ensure an uniform view of the Cloud services. In particular it uses: Cloud Computational Independent Model (CCIM) Cloud Platform Independent Model (CPIM). Cloud Computational Independent Model (CCIM) describes what the system is expected to, hiding all the technical details related to the implementation of the system CCIM Cloud Computational Independent Model describes what the system is expected to do, hiding all the technical details related to the implementation of the system CloudML, MODAClouds IDE Use of models for: Usage Service definition Service orchestration Requirements Data Strong influence on the design of applications: reducing the development time D2.2, D4.2.1, D5.1 from Plug-ins in Modelio Public Final Version 1.0, 31/03/

20 CPIM Cloud Platform Independent Model (CPIM) describes views of the systems in a platform independent manner so that it can be mapped to several platforms at the CPSM levels CloudML, MODAClouds IDE, CPSM Use models for: Data Design alternatives & deployment QoS Monitoring Resource Strong influence on the applications life-cycle: reducing the development & deployment time D2.2, D4.2.1, D4.3, D5.1 from Plug-ins in Modelio CPSM Cloud Platform Specific Model (CPSM) refines the CPIM with technical details required for specifying how the system can use a specific platform CPIM, CloudML, MODAClouds IDE Use models for: Data Design alternatives & deployment QoS Monitoring Resource Strong influence on the applications life-cycle: reduces the development and deployment time and enable the control of the deployed application D2.2, D4.2.1, D4.3.1, D5.1 from Plug-ins in Modelio Public Final version 1.0, 31/03/

21 4.2.3 Avoid lock-in MODAClouds is building an integrated solution for supporting the development, deployment and runtime control of applications and services using services from multiple Cloud providers, being therefore a Multi-Cloud solution. Beyond the abstract level that is used via the model-driven approach, a key component of the solution in avoiding the vendor lock-in is the Execution platform. MODAClouds Runtime Platform Enables the consumption of IaaS and PaaS services offered by major Cloud providers Monitoring platform, Execution platform, Self-adaptation platform, CloudML Use for the deployment of component based applications Portability of component based applications D6.1 and D3.2.1 from Execution environments supported by CloudML, mosaic and Cloud4SOA Public Final Version 1.0, 31/03/

22 4.3 Transform Roadmap MODAClouds white paper Underpinning the Leap to DevOps Movement on Cloud Environment is arguing for the adaption of the model driven approach to support the for the current DevOp style of work. Support for DevOps MODAClouds is part of the DevOps movement as it provides a set of tools that facilitates adoption of business and technical requirements very early into the application lifecycle and enables operations monitoring and portability capabilities to comply with expected operational quality levels All other BP from this document CloudML language enables the specification of the requirements at design time. Dev and Ops resources are designed together to meet the production environment conditions and taking into account the business requirements The set of tools are enabling monitoring and portability, that prevent double handling of information between development and operations while providing application and performance assurance for testing or production scenarios The semi-automated code generation for multiple Clouds enables faster deployment MODAClouds is to propose tools that take the requirements from Dev teams, apply them on the Ops side, and then close the circle by feeding information and insights extracted from running environments back to Dev side to improve or correct cloud applications Public Final version 1.0, 31/03/

23 4.3.2 Consulting The scope of business models that exploit EC project outputs is referring to a large range from service brokerage to open source technology for a wide community. At the current early stage of the MODAClouds project, the following business scenarios are investigated: 1. Technology as Open Source to the Community: deliver the core results as Open Source Software to the community. 2. Standalone Consulting Services over Open Source Approach: provide consulting services in using MODAClouds technology and knowledge, customization, installation, service, and support to third party organizations. 3. Technology Out of the Box: sell MODAClouds services (technology and knowledge) to customers to address their identified problems. 4. Recommendation-as-a-Service: refers to MODAClouds Decision Support System to be offered commercially. 5. MODAClouds-as-a-Service Broker: a new legal entity with the participation of project partners brokers the services of the partners. 6. Scientific Community Influence: use MODAClouds results to create influence on areas related to the project scope advancing the State of The Art and transferring knowledge and assets to the scientific community. Consulting services are already guaranteed to be provided after the end of the project. MODAClouds consulting services Selling software consulting services around the results by offering standard open source services after the project ends by industrial partners or by a new entity (installation, customization, support, training, etc.) All other BP from this document Sales and marketing to reach customers Promotion on Open Source Communities for uptake and development of MODAClouds Ready Adapters for Service Providers Direct revenue stream coming from consultancy services for the consortium partners Indirect non-commercial revenue streams could be generated by branding or influence benefits of open source Third party sponsorship or advertising could provide yet more indirect revenue D from Public Final Version 1.0, 31/03/

24 4.3.3 Scenarios MODAClouds team is building four case studies as proof-of-concept related to the usefulness of the modeldriven engineering approach for application development, deployment and run-time control on multiple Clouds: 1. Project management server 2. Business process modelling system 3. Palliative care application 4. Smart city urban safety planner The usage that these case studies will make of the tools developed in the project will allow us to define best practices in the final version of this document. At this point in time we use the best practice template to describe the case studies. Project management server Designing multi-cloud application Transitioning from desktop application to cloud application MODAClouds IDE, CloudML, Monitoring Platform, Run-time platform Use MODAClouds IDE to design the application at CCIM, CPIM and CPSM levels Separate cloud independent and cloud dependent parts of the application design Separate cloud independent and independent QoS constraints thanks to the monitoring constraints Identify how deployment changes from one provider to another. Simplify the evaluation of the costs of migrating to and between clouds Design of the case study in D8.2.1 Research papers of SOFTEAM in the publication repository: Modelio Initial design of the case study and evaluation of candidate cloud providers based on MODACloudML monitoring constraints and rules Public Final version 1.0, 31/03/

25 Business process modeling system Migration of legacy application re-deployed as a SaaS DSS, risk, and utility analysis to select the best IaaS Model based, cloud provider agnostic deployment to IaaS IaaS to IaaS migration Rule-based real-time application monitoring in multiple clouds DSS, MODAClouds IDE, Run-time platform, Data package, Monitoring platform Uses models and tools: MODAClouds Decision Support System for risk analysis and provider selection Model deployments in a cloud provider independent way using the MODAClouds IDE Deploy application artifacts and set op monitoring policies semi-automatically based on deployment models and monitoring rules provided through the MODAClouds IDE Predict performance based on usage models defined using the MODAClouds IDE Synchronize and migrate data at runtime Monitori and measure multiple cloud deployments Showcase for the migration of an existing web based application towards an SaaS solution in the Cloud, easy migration of the application stack from one IaaS to another and real-time monitoring in multiple clouds Design of the case study in D8.3.1 Research papers of BOC in the publication repository: Flexiant Flexiscale, Amazon AWS Public Final Version 1.0, 31/03/

26 Palliative care application Develop new application services Data managed on a private IaaS Heterogeneous environments: virtual desktops, application logic; hybrid Clouds (private IaaS, multiple public PaaS) Validate activities for filling the gap between runtime and design time DSS, MODAClouds IDE, Run-time platform, Data package, Monitoring platform Uses models and tools: Cost analysis Functional modeling QoS design tools Adaptive policies Data replication Showcase for the designing an application following the model driven approach, data management in private IaaS and hybrid Clouds, as well as adaptivity at run-time Research papers of ATOS in the publication repository: Cloud4SOA Public Final version 1.0, 31/03/

27 Smart city urban safety planner Develop a new application High performance, scalability, and availability requirements Data design and run-time management PaaS to PaaS migration DSS, MODAClouds IDE, Run-time platform Uses models and tools: Cost analysis Functional modeling QoS design tool Adaptive policies Showcase for Designing a highly available system portable on multi-cloud environment Auto-scaling system to a dynamic load Deploying a distributed Complex Event Processing system on the cloud Research papers of SIEMENS in the publication repository: MODAClouds IDE Public Final Version 1.0, 31/03/

28 4.4 Practice Open Cloud Ecosystem MODAClouds is producing open-source codes, most of them under Apache 2.0 licence. Open source codes In order to promote the usage of Cloud services and tools, several MODAClouds components are released with open-source code licenses MODAClouds IDE, CloudML, Monitoring platform, Execution platform Use of well-known open-source code repositories Re-use of the open-source components of interest for the developers communities GitHub, BitBucket, GoogleCode, SourceForge Public Final version 1.0, 31/03/

29 4.4.2 Cloud Migration Management MODAClouds is tackling with the issue of data migration from a source storage service to a target one. Data partition and replication MODAClouds (i) investigates the set of data models a cloud application developer can select depending on the application he/she is developing, (ii) describes the approach to map such models into the target data storage services, which typically offer a data model that is less expressive but allows for a high efficiency, and (iii) discusses about the issue of data migration from a source storage service to a target one. Execution platform Use data models: CCIM data model in which the date is described in terms of the entity-relationship model CPIM data model as logical data model that aims at describing data structures taking into account the classes of data storage (like NoSQL, DFS, blob) in which the date is described using a graph data model (GDM), a hierarchical model (HDM), or a flat data model (FDM). CPSM data model that defines concepts strongly connected to the cloud providers services in which the data are described using a relational data model, a key-value data model, a column-based data model and so on Use migration approaches if the same CPIM model is used (e.g. GDM to GDM): switch-off/switch-on runtime If using the same data models at Cloud provider independent layer, the data can be migrated from one Cloud to another D4.4.1 from Marija framework Public Final Version 1.0, 31/03/