Declarative and Interoperable Overlay Networks, Applications to Systems of Systems (DIONASYS)

DIONASYS项目发生在一个快速变化的计算世界的背景下。大量异构资源相互连接。在智慧城市、应急响应系统、医疗保健系统、能源系统和智能电网等领域都是如此。为如此大型的相互连接的异构系统构建应用程序是一项复杂的任务。它需要考虑最终用户和应用程序需求，例如服务级别和端到端功能。它还需要考虑底层及其约束。这些限制可能因系统而异：连接性、移动性、能源方面和可用性是常见的例子。在这个项目中，我们从环境和地球观测系统的特定领域获得灵感，并将通过这些高度异构的分布式系统来展示我们的成果。这样的系统实际上是更准确的分布式系统，具有包括广泛的无线传感器网络在内的底层子系统。这些传感器覆盖不同地理位置的不同环境参数（例如与土壤、水、污染物有关的环境参数），并以不同的尺度测量属性。然后将数据馈送到云基础设施中，在大多数情况下，云基础设施是联合和混合的。在此基础上，还需要在移动平台上访问数据和其他环境资源（例如，预测模型输出的可视化）。覆盖网络提供了一个完善的解决方案来支持分布式系统中的低级原语。然而，在复杂的分布式系统的系统中，联合异构分布式系统并支持不同类型的覆盖，固有地出现了高度的复杂性。我们建议利用分布式系统、语言、网络和中间件领域公认的研究团队的一套独特的能力和技能来解决可组合和可互操作的覆盖设计和适应机制的需求，并简化复杂、多样和异构分布式系统的应用程序设计。我们提出了一种生成方法来设计和实现覆盖。我们将从全局抽象的角度构建一种特定于领域的语言来指定覆盖。而不是指定细粒度的相互作用，覆盖设计人员将能够指定所需覆盖的全局特征，从结构和功能的角度来看。这些描述将被编译成代码，这些代码将被运行在所考虑的分布式系统的异构节点上的动态运行时所支持。生成的代码将利用基于八卦范例的自组织分布式协议。这样的协议允许从节点之间简单的局部交互中自然地表达全局行为。它们的灵活性也是解决适应性和可组合性需求的关键。特定于领域的语言将允许指定跨多个覆盖层的适应性和组合约束和需求。语言和运行时将同时支持多个覆盖层之间的互操作性，允许动态地水平插入覆盖层（作为覆盖多个网络的系统的系统，并提供集成的端到端功能），并允许垂直堆叠覆盖层，在堆栈增长时提供增加的抽象和更丰富的服务和功能。我们将在环境和地球观测系统的背景下，利用水平和垂直的可组合性和对异构系统系统的支持。

The DIONASYS project takes place in the context of a fast-changing computing world. Large amounts of heterogeneous resources get interconnected. This is true for example in areas such as smart cities, emergency response systems, healthcare systems, energy systems and smart grids. Building applications for such large inter-connected heterogeneous systems is a complex task. It requires taking into account end-users and applications requirements, such as service levels and end-to-end functionalities. It requires as well taking into account underlying layers and their constraints. These constraints can differ largely from one system to the other: connectivity, mobility, energy aspects and availability are prevalent examples.In this project, we take inspiration from, and will demonstrate our results through, the specific domain of environmental and earth observation systems, which are highly heterogeneous distributed systems. Such systems are actually more accurately distributed systems of systems with underlying subsystems including a wide range of wireless sensor networks. These sensors cover different environmental parameters (e.g. related to soils, water, pollutants) at different geographical locations and measure properties at different scales. The data is then fed into cloud infrastructures that in most cases are federated and hybrid. Layered on top of this there is also a need to access data and other environmental resources on the move on mobile platforms (e.g. visualizations of outputs from prediction models).Overlay networks offer a well-established solution to underpin low-level primitives in distributed systems. However, in complex distributed systems-of-systems, that federate heterogeneous distributed systems and support different kinds of overlays, high levels of complexity inherently occur. We propose to leverage a unique set of competencies and skills of well-recognized research teams in distributed systems, languages, networks and middleware to address the need for composable and interoperable overlay design and adaptation mechanisms and simplify the design of applications for complex, diverse and heterogeneous distributed systems.We propose a generative approach to overlay design and implementation. We will build a domain-specific language for specifying overlays, from a global-scale abstraction perspective. Instead of specifying fine-grain interactions, overlay designers will be able to specify global characteristics of the required overlays, both from a structural and functional point of view. The descriptions will be compiled to code that will be supported by a dynamic runtime running on the heterogeneous nodes of the considered distributed system(s). The generated code will take advantage of self-organizing distributed protocols built upon the gossip-based paradigm. Such protocols allow naturally expressing global behaviours from simple local interactions between nodes. Their flexibility will also be key to addressing the need for adaptation and composability.The domain-specific language will allow specifying adaptation and composition constraints and requirements across multiple overlays. Interoperability between multiple overlays will be supported both by the language and runtime, allowing to dynamically plug overlays horizontally-as systems of systems ranging multiple networks and providing integrated end-to-end functionalities-and vertically, by allowing stacking overlays, providing increased abstraction and richer services and functionalities when growing up the stack. We will exploit the horizontal and vertical composability and support for systems of heterogeneous systems in the context of environmental and earth observation systems.