UML-based Cloud Application Modeling with Libraries, Profiles, and Templates? Alexander Bergmayr, Javier Troya, Patrick Neubauer, Manuel Wimmer, and Gerti Kappel Business Informatics Group, Vienna University of Technology, Austria {bergmayr,troya,neubauer,wimmer,kappel}@big.tuwien.ac.at Abstract. Recently, several cloud modeling approaches have emerged. They ad- dress the diversity of cloud environments by introducing a considerable set of modeling concepts in terms of novel domain-specific languages. At the same time, general-purpose languages, such as UML, provide modeling concepts to represent software, platform and infrastructure artifacts from different viewpoints where the deployment view is of particular relevance for specifying the distribu- tion of application components on the targeted cloud environments. However, the generic nature of UML’s deployment language calls for a cloud-specific exten- sion to capture the plethora of cloud provider offerings at the modeling level. In this paper, we propose the Cloud Application Modeling Language (CAML) to fa- cilitate expressing cloud-based deployments directly in UML, which is especially beneficial for migration scenarios where reverse-engineered UML models are tai- lored towards a selected cloud environment. We discuss CAML’s realization as a UML internal language that is based on a model library for expressing deploy- ment topologies and a set of profiles for wiring them with cloud provider offer- ings. Finally, we report on the use of UML templates to contribute application deployments as reusable blueprints and identify conceptual mappings between CAML and the recently standardized TOSCA. Keywords: Cloud Computing, Model-Driven Engineering (MDE), Cloud Mod- eling, UML, Language Engineering 1 Introduction Cloud computing has recently emerged as a new possibility how software can be made available to clients as a service. For software vendors, this is appealing as cloud envi- ronments [3] have the benefit of low upfront costs compared to a traditional on-premise solution and operational costs that scale with the provisioning and releasing of cloud offerings. They may range from low-level infrastructure elements, such as raw com- puting nodes, over higher level platforms, such as a Java execution environment on top of a cloud infrastructure, to ready-to-use software deployed on a platform. As a result, current cloud environments are diverse in nature and show various levels of vir- tualization they operate on. Recent cloud modeling approaches already capture a con- siderable set of domain-specific concepts to support different scenarios: description of ? This work is co-funded by the European Commission under the ICT Policy Support Pro- gramme, grant no. 317859. cloud-based applications [17] and their deployments [5, 13, 21], optimization of such deployments [14, 18], provisioning of cloud resources [10], or automating the scalabil- ity of cloud environments [9,11]. At the same time, general-purpose languages, such as UML, provide modeling concepts to represent software, platform and infrastructure ar- tifacts from different viewpoints. Hence, providing extensions to UML that satisfy cur- rent cloud modeling requirements appears beneficial, especially when cloud-oriented migration scenarios [4] need to be supported where reverse-engineered UML models are tailored towards a selected cloud environment. However, to date, effective UML-based support for modeling cloud application de- ployments that are wired with cloud provider offerings is still missing. As a result, on-premise deployments expressed in UML can hardly be turned into cloud-based de- ployments without neglecting the intended usage of UML. In the ARTIST project [4], we are particularly confronted with this problem as we work towards a model-driven engineering approach for modernizing applications by novel cloud offerings, which in- volves deploying them or at least some of their components on a cloud environment. Ideally, the design choices of a cloud-based deployment are expressed at the modeling level, which calls for an appropriate language support in the light of UML. While in this way, not only the full expressive power of UML can be exploited, also a seamless integration of cloud-specific models into existing UML models is ensured. In this paper, we propose the Cloud Application Modeling Language (CAML) [7] to enable representing cloud-based deployment topologies directly in UML and refin- ing them with cloud offerings captured by dedicated UML profiles. Thereby, a clear separation is achieved between cloud-provider independent and cloud-provider specific deployment models [1], which is in accordance with the PIM/PSM concept. In our case, the “platform” refers to the cloud provider. We developed profiles for two major cloud providers1 and integrated them into a common cloud profile. Inspired from common cloud computing literature [2, 3, 12], recent cloud modeling approaches [5, 9, 15, 17, 18, 21] and cloud programming approaches2 , we developed CAML’s model library that facilitates developing base deployment topologies to which cloud offering profiles are applied. The benefits of realizing CAML as an internal language of UML are threefold: (i) UML provides a rich base language for the deployment viewpoint, (ii) “cloudify- ing” UML models is facilitated without the need to re-model existing applications, and (iii) profiles in UML allow hiding details of cloud provider offerings from models and dynamically switching between them by (un-/re-)applying respective cloud provider profiles. We motivate the practical value of CAML by means of a deployment scenario in Section 2. In Section 3, we give the design rationale of CAML and provide insights into its model library and the covered UML profiles whereas in Section 4, we discuss the employment of UML templates as reusable deployment blueprints. The operational- ization of CAML by means of a mapping to the recently accepted TOSCA standard is dedicated to Section 5. Finally, in Section 6 we discuss work related to CAML before we conclude in Section 7. 1 Amazon AWS: http://aws.amazon.com and Google Cloud Platform: http://cloud.google.com 2 Deltacloud: https://deltacloud.apache.org and jclouds: http://jclouds.apache.org 2 Motivating Deployment Scenario To motivate the benefits of employing UML as the host language for realizing CAML, we give an overview of UML’s structural viewpoints that support representing appli- cation deployments by means of a reference application3 of the ARTIST project. We take the viewpoint of the application components and their deployment. Figure 1a de- picts some components of our application, an excerpt of their realizing classes and the manifestation of these components by deployable artifacts. A possible on-premise de- ployment for them is presented in Figure 1b. It covers instances of the two deployable artifacts and connects them to a Java-based middleware and a relational DBMS, which are in turn deployed onto a node with specified (virtual) machine characteristics. The model elements of the deployment are instances of the custom types defined in the component viewpoint (see Figure 1a) and the deployment viewpoint (see Figure 1c), respectively. With the emergence of cloud offerings and the demand to exploit them, deployment models need to be expressive enough to capture such offerings. This is ex- actly the idea of CAML. Because it is realized in terms of lightweight extensions to UML, CAML models are applicable to UML models and so to our modeled reference application as depicted in Figure 1. In Sections 3 and 4, we present cloud-based deploy- ments for our reference application. Petstore Components «use» «component» «component» «component» PetstoreWeb PetstoreService PetstoreDomain «class» «use» «class» «use» «class» «class» ShoppingCart OrderService Order OrderLine - order:Order + createOrder():Order - orderId:long - lineId:long 1..* - orderService:OrderService + findOrder(in id long):Order «manifestation» «manifestation» «manifestation» «artifact» «artifact» PetstoreBusiness PetstoreData «package import» (a) Component Viewpoint On-premise Petstore Deployment «ModelLibrary»Web Deployment Library :PetstoreBusiness :PetstoreData «ExecutionEnvironment» «Enumeration» «ExecutionEnvironment» ApplicationContainer ContainerKind Datastore «deploy» «deploy» «deploy» type:ContainerKind [1] JEE type:DatastoreKind [1] :ApplicationContainer :Datastore RubyOnRails container=JEE type=relational «Node» «Enumeration» «Enumeration» «deploy» OnPremiseNode OSKind DatastoreKind «package memory:Real [0..1] Linux Relational :OnPremiseNode import» CPU:Real [0..1] Windows DocumentOriented memory=2 localDisk:Real [0..1] CPU=1.7 «deploy» operatingSystem:OSKind [0..1] localDisk=4 operatingSystem=Linux (b) Deployment Viewpoint at Instance Level (c) Deployment Viewpoint at Type Level Fig. 1: CAML Use-Case 3 Cloud Application Modeling With CAML, we propose lightweight extensions to UML for modeling cloud application deployments that are seamlessly applicable to UML models, such as component mod- els, typically created throughout software modeling activities. The intended purpose of 3 It is based on the Java Petstore: http://www.oracle.com/technetwork/java/index-136650.html CAML is to express deployment topologies by common cloud modeling concepts and to enable the wiring of such models with concrete cloud provider offerings. This wiring is achieved by applying a dedicated CAML Profile to a deployment model expressed in terms of the CAML Library. As a result, a clear separation between cloud-provider independent and cloud-provider specific models is achieved. Selecting cloud provider offerings at the modeling level for a concrete deployment becomes a matter of applying the respective stereotypes. The overall set of stereotypes encompass the possible design choices provided by CAML regarding cloud provider offerings. 3.1 Model Library for Cloud Deployment Topologies As presented in Figure 2, the CAML Library is built around the concept of cloud offer- ing. It is considered as a virtual resource that is expected to be supported by a cloud environment once the wiring with a concrete cloud offering has been performed. More specifically, three offering types capture common cloud environment capabilities. A cloud node provides compute capacity and operates at a certain level of virtualiza- tion [3]. From an infrastructure-level perspective, cloud nodes come with an operat- ing system, while when turning this perspective to the platform level they also provide middleware, such as a web server and an application container. In case of the latter, the platform is fully managed by a cloud environment. With dedicated scalability strategies, the elastic nature of a cloud environment is managed. For instance, cloud nodes can automatically be acquired depending on the number of incoming requests. Clearly, ac- quiring and releasing cloud nodes can also be manually controlled. The second offering refers to the cloud storage capabilities of cloud environments which provide diverse so- lutions for structuring application data [12] and increasing their availability by relaxing consistency [22]. Finally, a cloud service is considered as a ready-to-use cloud offering that is provisioned and managed by a cloud provider. For instance, a load balancer that distributes requests to cloud nodes is an infrastructure-related cloud service, while a task queue for long running processes is a platform-related cloud service. To represent offering-to-offering connections, communication channels are employed while cloud configurations enable modifying the assumed conventions of a cloud environment. For instance, an automatic scaling strategy can be configured with boundaries of minimum and maximum running cloud nodes. Generally, instantiated elements of the CAML Li- brary are refined to concrete cloud provider offerings via dedicated stereotypes. «ModelLibrary» CloudLibrary Cloud Library «CommunicationPath» CommunicationChannel channelSource «Enumeration» «Enumeration» «Association» [*] VirtualizationKind ScalingStrategy «DeploymentSpecification» OfferingConfiguration «Class» infrastructure automatic CloudConfiguration CloudOffering [*] channel platform manual [*] offering configuration [*] Target «Enumeration» «Enumeration» StructureKind ConsistencyKind «Node» «ExecutionEnvironment» «Artifact» Block strict CloudNode CloudStorage CloudService Blob eventual virtualization:VirtualizationKind [1] dataStructure:StructureKind [1] Relational scaling:ScalingStrategy [1] consistency:ConsistencyKind [1] KeyValue Fig. 2: Cloud Library of CAML 3.2 Profiles for Cloud-Provider Specific Deployments With CAML Profiles, we provide a set of UML stereotypes that enable wiring cloud deployment topologies with concrete offerings of cloud providers. Basically, a stereo- type embodies a concrete offering at the modeling level and captures its features in terms of properties. Figure 3 presents some stereotypes specific to the cloud offerings of the Google App Engine (GAE) and Amazon AWS. Common cloud offerings that are shared by both providers are lifted to the common cloud profile. Considering instance types, they are supposed to be applied to cloud nodes to wire them to a concrete cloud offering, such as a “Frontend Instance” (e.g., GAEF1) that hosts a Java-based middle- ware managed by Google’s App Engine. In turn, cloud offerings are refined by what we call meta-profiles. With the notion of meta-profiles, we facilitate refining them with technical-related details, such as the performance of instance types, and business-related information [8], like the costs of cloud offerings. CommonCloudProfile «metaclass» InstanceSpecification RequiresCloudNodeClassifier {{OCL} self.base_InstanceSpecification «Stereotype» .classifier->any(e|e.oclisTypeOf(CloudNode)) InstanceType ->notEmpty()} «package import» «package import» GAECloudProfile AWSCloudProfile Cloud Provider Profiles «Enumeration» «Enumeration» «Stereotype» «Enumeration» MiddlewareKind OSKind AWSInstanceType RegionKind Java «Stereotype» RHEL operatingSystem:OSKind [1] US_EAST GAEInstanceType Go SLES region:RegionKind [1] EU PHP middleware:MiddlewareKind [1] Windows availabilityZone:String [0..1] ASIA_Singapore Python «generalPurpose,runningCosts» «storageOptimized,runningCosts» «generalPurpose,runningCosts» «generalPurpose,runningCosts» «Stereotype» «Stereotype» «Stereotype» «Stereotype» AWSM3Medium AWSC3Large GAEF1 GAEF4 «GeneralPurpose» «StorageOptimized» «GeneralPurpose» «GeneralPurpose» memory=3.75 memory=3.75 memory=0.128 memory=0.512 virtualCores=1 virtualCores=2 CPU=0.6 CPU=2.4 localDisk=4 localDisk=32 «RunningCosts» «RunningCosts» «RunningCosts» «RunningCosts» value=0.05 value=0.20 value=0.70 value=0.105 currency=USD currency=USD currency=USD currency=USD metric=Hour metric=Hour metric=Hour metric=Hour «profile application» «profile application» «profile application» PricingProfile PerformanceProfile «metaclass» «metaclass» Stereotype Stereotype «Stereotype» «Enumeration» «Stereotype» Meta-Profiles CostComponent CurrencyKind InstanceTypeCharacteristics value:Real [1] EUR memory:Real [0..1] currency:CurrencyKind [1] USD virtualCores:Integer [0..1] metric:MetricKind [1] CPU:Real [0..1] «Enumeration» localDisk:Real [0..1] MetricKind Hour «Stereotype» «Stereotype» GB «Stereotype» «Stereotype» RunningCosts BaseCosts Operation GeneralPurpose StorageOptimized «profile application» Fig. 3: CAML Profiles and Meta-Profiles 3.3 CAML By-Example To demonstrate how CAML is applied, Figure 4 presents a possible deployment topol- ogy and refinement towards a GAE-based cloud deployment of our introduced use case (cf. CAML Figure 1). In a first step, we modeled the deployment topology. It consists of two au- TOSCA tomatically :PetstoreDatascaled cloud nodes and a key-value cloudDataTier myPetstoreData (DataTier) storage for managing the applica- deploy tion data in an eventually consistent way. As the cloud nodes are specified as platform- CloudNode «deploy» (deploy) Legend level offering, weCAML2 directly deployed the application TOSCA components virtualization:VirtualizationKind [1] onto them. Then, in NodeTemplate «gAEF4» myCloudNode scaling:ScalingStrategy [1] a second:CloudNodestep, we applied the GAE (GAEF4)profile and the respective stereotypes RelationshipTemplate to refine the deployment model towards virtualization=platform virtualization=platform scaling=Auto concrete cloud offerings GAEF4 scaling=Auto provided by the GAE.NodeType As a result, RelationshipType the modeled «GAEF4» cloud nodes refer to the F1 and F4 instance types Middleware=Java middleware:MiddlewareKind [1] that host a Java-based middleware=Java DerivedFrom middleware. The configurationTemplates attached to these cloudTypes nodes constrains the maximum number of idle cloud nodes. Finally, GAE’s key-value datastore is employed for the required cloud storage capabilities. «ModelLibrary»CloudLibrary «ModelLibrary»CloudLibrary «Profile»GAECloudProfile «package import» «package import» «profile application» Petstore Deployment Topology GAE-based Petstore Deployment :PetstoreBusiness :PetstoreData :PetstoreBusiness :PetstoreData «deploy» «deploy» «deploy» «deploy» «gAEF1» «gAEF4» :CloudNode :CloudNode :CloudNode :CloudNode virtualization=platform virtualization=platform virtualization=platform virtualization=platform GAE-based scaling=Auto scaling=Auto scaling=Auto scaling=Auto Refinement «GAEF1» «GAEF4» middleware=Java middleware=Java «appEngineDatastore» «autoScaledConfiguration» «appEngineDatastore» :CloudStorage :CloudConfiguration :CloudStorage dataStructure=KeyValue «AutoScaledConfiguration» dataStructure=KeyValue consistency=Eventual maximumIdleInstances=3 consistency=Eventual (a) Deployment Topology of CAML Use-Case (b) GAE-based Deployment of CAML Use-Case Fig. 4: Reference Application deployed onto Google App Engine 3.4 Prototypical Implementation To show the feasibility of CAML, we have implemented an Eclipse-based prototype, which exploits extension points. In this way, developers can directly use CAML in Eclipse tools, such as Papyrus4 , or access its library and profiles in terms of a resource, which is helpful for the development of transformations. CAML together with all arti- facts used in this paper are publicly available at our project web site [7]. In addition, together with our industrial partner SparxSystems, we have also implemented a first version of CAML for Enterprise Architect5 . This provides first evidence that our pro- posed approach for developing a UML internal cloud modeling language based on a library and profiles is feasible and current modeling tools with UML support provide the necessary features to support CAML models. 4 Papyrus: http://www.eclipse.org/papyrus 5 Enterprise Architect: http://www.sparxsystems.at 4 Reusable Deployment Blueprints as UML Templates As CAML is based on UML, its reuse mechanisms can be applied for cloud applica- tion deployments. This is particularly useful for providing frequently occurring deploy- ment patterns as predefined UML templates. To show their usefulness and give first evidence of CAML’s expressivity, we developed 10 templates as reusable deployment blueprints, most of them are based on Amazon’s best practices6 . We modeled their in- herent topology with CAML’s cloud library and refined them with stereotypes from the cloud profile dedicated to Amazon. The developed blueprints are available at our project website [7]. To demonstrate the use of a blueprint, we show how our reference application is bound to a template, which refers in our case to a 2-tier web architec- ture [12]. To reuse the predefined template, the deployable artifacts need to be bound to the template parameters. Figure 5 depicts the component viewpoint of our reference application and the respective CAML template. It consists of two cloud nodes that refer to the “M3Medium” offering of Amazon. Their location is required to be in Europe while the operation system needs to be Linux. For reliability reasons, they are placed in different availability zones. Requests that arrive at the cloud nodes are first handled by a load balancing service, which enables a higher fault tolerance of the application. The number of running cloud nodes is automatically managed by Amazon as expressed by the scalability strategy. Only the minimum number of running cloud nodes and their adjustment is configured. Both cloud nodes are connected to a cloud storage that in turn is replicated to improve data availability. Finally, as Amazon cloud nodes operate at the infrastructure level, the required middleware for our reference application is defined. In fact, we directly reused it from the on-premise deployment given in Figure 1. 6 Amazon Architecture Center: https://aws.amazon.com/architecture AWS-based Web Deployment BusinessTier:Artifact, DataTier:Artifact «elasticLoadBalancing» «artifact» «ModelLibrary» :CloudService BusinessTier CloudLibrary :BusinessTier :DataTier «artifact» DataTier «deploy» «deploy» «package import» :WebContainer «deploy» «deploy» container=JEE6 «ModelLibrary» «aWSM3Medium» «aWSM3Medium» WebDeploymentLibrary :CloudNode :CloudNode «autoScalingGroup» virtualization=infrastructure :CloudConfiguration virtualization=infrastructure scaling=Auto «AutoScalingGroup» scaling=Auto «package import» «AWSM3Medium» minimumInstances=3 «AWSM3Medium» operatingSystem=Linux adjustment=1 operatingSystem=Linux «Profile» region=EU region=EU AWSCloudProfile availabilityZone=“A” availabilityZone=“B” «rDSMaster» «rDSSlave» :CloudStorage :CloudStorage «profile application» dataStructure=Relational dataStructure=Relational consistency=strict consistency=strict «bind» Petstore Components «use» «component» «component» «component» PetstoreWeb «use» PetstoreService «use» PetstoreDomain «manifestation» «manifestation» «manifestation» «artifact» «artifact» PetstoreBusiness PetstoreData Fig. 5: Reusable Deployment Template for AWS 5 Interoperability between CAML and TOSCA One major aspect in model-based engineering is to place models as first-class entities in the engineering process. Ideally, they should be turned into executable or interpretable artifacts. Regarding the deployment viewpoint, it appears desirable to translate the re- spective models into descriptors and scripts that are passed to provisioning engines for cloud environments. For instance, a GAE-based deployment requires specific descrip- tors for defining the assignment of application modules to a concrete instance type. This assignment can certainly be derived from a CAML model. At the same time, there are ongoing efforts in standardizing the representation of cloud-based application de- ployments. The recently accepted TOSCA standard aims at supporting portable cloud applications. With the notion of management plans, emerging TOSCA-compliant en- gines are capable to interpret such deployment topologies and initiate the provision- ing of defined service templates [5]. Clearly, this is also of practical value for CAML models. For that reason, we present an initial mapping between CAML and a subset of TOSCA. Generally, in TOSCA, two modeling concepts are prevalent: template and type. Templates embody the elements of a deployment topology while types expose the properties and relationships for which concrete values are provided by templates. In this sense, types are considered as reusable entities that can inherit from each other. Figure 6 depicts a concrete TOSCA model expressed in Vino4TOSCA [6] for an excerpt of our GAE-based application deployment (cf. Figure 4). To represent the TOSCA template for the stereotyped CAML cloud node, the pertinent TOSCA types need to be created: “CloudNode” and “GAEF1”. The latter is derived from the former as in TOSCA a template can only have a single type. Similarly, the deployed application component is represented by a TOSCA template. Finally, the deployment relationship type is required for connecting the deployed application component to the cloud node at the template level. CAML TOSCA :PetstoreData myPetstoreData DataTier (DataTier) deploy «deploy» (deploy) CloudNode Legend CAML2 virtualization:VirtualizationKind [1] TOSCA scaling:ScalingStrategy [1] NodeTemplate «gAEF4» myCloudNode :CloudNode (GAEF4) RelationshipTemplate virtualization=platform virtualization=platform NodeType scaling=Auto scaling=Auto GAEF4 RelationshipType «GAEF4» Middleware=Java middleware:MiddlewareKind [1] middleware=Java DerivedFrom Templates Types Fig. 6: Mapping between CAML and TOSCA «ModelLibrary»CloudLibrary «ModelLibrary»CloudLibrary «Profile»GAECloudProfile «package import» «package import» 6Petstore Related Work Deployment Topology GAE-based Petstore Deployment «profile application» :PetstoreBusiness :PetstoreData :PetstoreBusiness :PetstoreData Cloud modeling approaches «deploy» «deploy» with the purpose of«deploy» achieving «gAEF1» the wiring «deploy»of applications «gAEF4» with concrete :CloudNode cloud offerings :CloudNodeare most closely related to CAML. Modeling :CloudNode :CloudNode concepts of virtualization=platform virtualization=platform virtualization=platform virtualization=platform these scaling=Auto 9, 15, 18, 21] areGAE-based approaches [5,scaling=Auto reflected Refinement by CAML scaling=Auto on a level of abstraction scaling=Auto that «GAEF1» «GAEF4» facilitates to represent design decisions for cloud-based middleware=Javaapplicationmiddleware=Java deployments. As a «appEngineDatastore» «autoScaledConfiguration» «appEngineDatastore» :CloudStorage :CloudConfiguration :CloudStorage dataStructure=KeyValue «AutoScaledConfiguration» dataStructure=KeyValue consistency=Eventual maximumIdleInstances=3 consistency=Eventual (a) Deployment Topology of CAML Use-Case (b) GAE-based Deployment of CAML Use-Case result, modeling concepts of these approaches, e.g., required to achieve the optimization of an application deployment (cf., [15]) or to express elasticity rules (cf., [9]), are not completely captured by CAML. However, CAML enables expressing cloud application deployments that are seamlessly applicable on UML models usually created throughout software modeling activities as it is realized as a UML internal language. As a result, well-connected modeling views on cloud applications from a cloud-provider indepen- dent perspective as well as a cloud-provider specific perspective are supported. The re- finement of modeling views is enabled by profiles for cloud providers. This additional typing dimension provided by such profiles and the exploitation of a multi-viewpoint language to realize CAML differentiates it from existing cloud modeling approaches and the recently standardized TOSCA. To the best of our knowledge, the only approach providing cloud modeling sup- port within UML is MULTICLAPP [17]. It proposes a UML profile for the purpose of representing components that are expected to be deployed onto a cloud environment by applying cloud-provider independent stereotypes to them. Hence, these stereotypes do not support wiring components with cloud provider offerings, which is different to CAML as stereotypes are applied to achieve exactly that wiring. CloudML-UFPE [16] provides modeling concepts to represent cloud offerings con- nected with the internal resources of a cloud environment. Similarly to approaches [10, 11, 19], which propose modeling concepts to represent resources internally managed by a cloud environment, the focus is set on the cloud provider perspective. As a result, such modeling approaches support cloud providers to model their environments, which is out of the scope of CAML. Finally, it is worth mentioning that approaches, such as Deltacloud7 and jclouds8 , provide an abstraction layer on top of cloud-provider specific programming libraries. They can be considered as transformation targets for cloud modeling approaches to automate the provisioning of modeled application deployments. 7 Conclusion and Future Work We have presented CAML as a UML internal language based on a library, profiles, and templates. Currently, it is employed by the ARTIST project to model deployments of large applications used in practice. In this respect, cloud providers that operate at both infrastructure level and platform level are targeted. Although the realization and initial evaluation of CAML seems promising, several lines of future work need to be investi- gated. First, we aim for an automated maintenance of provider-specific profiles with, for instance, performance or pricing information based on web information extraction techniques. Second, we intend to provide a simulator for CAML to provide prediction about non-functional properties such as costs and performance. In this respect, we plan to explore how FUML can be employed to provide behavioral semantics for CAML in a similar way as we use it to define behavioral semantics for metamodels [20]. Finally, we aim for interoperability with current cloud modeling approaches by providing dedicated transformations or a UML profile. 7 https://deltacloud.apache.org 8 https://jclouds.apache.org References 1. Ardagna, D., Nitto, E.D., Mohagheghi, P., Mosser, S., Ballagny, C., D’Andria, F., Casale, G., Matthews, P., Nechifor, C.S., Petcu, D., Gericke, A., Sheridan, C.: MODAClouds: A Model-Driven Approach for the Design and Execution of Applications on Multiple Clouds. In: MISE Workshop (2012) 2. Armbrust, M., Fox, A., Griffith, R., Joseph, A.D., Katz, R.H., Konwinski, A., Lee, G., Pat- terson, D.A., Rabkin, A., Stoica, I., Zaharia, M.: A View of Cloud Computing. CACM 53(4) (2010) 3. Badger, M.L., Grance, T., Patt-Corner, R., Voas, J.M.: Cloud Computing Synopsis and Rec- ommendations. Tech. rep., NIST Computer Security Division (2012) 4. Bergmayr, A., Bruneliere, H., Cánovas Izquierdo, J.L., Gorroñogoitia, J., Kousiouris, G., Kyriazis, D., Langer, P., Menychtas, A., Orue-Echevarria Arrieta, L., Pezuela, C., Wimmer, M.: Migrating Legacy Software to the Cloud with ARTIST. In: CSMR (2013) 5. Binz, T., Breitenbücher, U., Kopp, O., Leymann, F.: TOSCA: Portable Automated Deploy- ment and Management of Cloud Applications. In: Advanced Web Services (2014) 6. Breitenbücher, U., Binz, T., Kopp, O., Leymann, F., Schumm, D.: Vino4TOSCA: A Visual Notation for Application Topologies based on TOSCA. In: OTM (2012) 7. CAML: Project Web Site (2014), http://code.google.com/a/eclipselabs.org/p/caml 8. Cardoso, J., Barros, A., May, N., Kylau, U.: Towards a Unified Service Description Language for the Internet of Services: Requirements and First Developments. In: SCC (2010) 9. Chapman, C., Emmerich, W., Márquez, F.G., Clayman, S., Gallis, A.: Software Architecture Definition for On-Demand Cloud Provisioning. Cluster Comput. 15 (2012) 10. Chatziprimou, K., Lano, K., Zschaler, S.: Towards a Meta-model of the Cloud Computing Resource Landscape. In: MODELSWARD (2013) 11. Dougherty, B., White, J., Schmidt, D.C.: Model-Driven Auto-Scaling of Green Cloud Com- puting Infrastructure. FGCS 28 (2011) 12. Fehling, C., Leymann, F., Retter, R., Schupeck, W., Arbitter, P.: Cloud Computing Patterns - Fundamentals to Design, Build, and Manage Cloud Applications. Springer (2014) 13. Ferry, N., Rossini, A., Chauvel, F., Morin, B., Solberg, A.: Towards Model-Driven Provision- ing, Deployment, Monitoring, and Adaptation of Multi-cloud Systems. In: CLOUD (2013) 14. Frey, S., Fittkau, F., Hasselbring, W.: Search-based Genetic Optimization for Deployment and Reconfiguration of Software in the Cloud. In: ICSE (2013) 15. Frey, S., Hasselbring, W.: The CloudMIG Approach: Model-Based Migration of Software Systems to Cloud-Optimized Applications. Advances in Software 4 (2011) 16. Gonçalves, G., Endo, P., Santos, M., Sadok, D., Kelner, J., Merlander, B., Mångs, J.E.: CloudML: An Integrated Language for Resource, Service and Request Description for D- Clouds. In: CloudCom (2011) 17. Guillén, J., Miranda, J., Murillo, J.M., Canal, C.: A UML Profile for Modeling Multicloud Applications. In: ESOCC (2013) 18. Leymann, F., Fehling, C., Mietzner, R., Nowak, A., Dustdar, S.: Moving Applications to the Cloud: An Approach Based on Application Model Enrichment. IJCIS 20(3) (2011) 19. Liu, D., Zic, J.: Cloud#: A Specification Language for Modeling Cloud. In: CLOUD (2011) 20. Mayerhofer, T., Langer, P., Wimmer, M., Kappel, G.: xMOF: Executable DSMLs based on fUML. In: SLE (2013) 21. Nguyen, D.K., Lelli, F., Taher, Y., Parkin, M., Papazoglou, M.P., van den Heuvel, W.J.: Blueprint Template Support for Engineering Cloud-Based Services. In: ServiceWave (2011) 22. Vogels, W.: Eventually consistent. CACM 52(1) (2009)