Discussion and Related Work

Having compared dSpace to specific IoT systems in §6, we now discuss the broader context of IoT.

Industry solutions. The IoT landscape today includes a wide range of solutions that one might view as taking one of three forms: device-centric, app-centric, or infrastructure-centric. Device-centric solutions are ones in which the vendor starts with their own device(s) and then offers optional add-on software and cloud services for these devices. Such vendors are typically device manufacturers such as Philips, LIFX, Geeni, BOSE, Dyson, and WYZE [5, 15, 21, 25, 35, 40, 49]. dSpace can integrate these device-centric solutions as we demonstrate in §6. We use the term app-centric to refer to solutions in which the vendor has built an application and they expose APIs that allow heterogeneous devices to integrate with this application or service. Such vendors are typically smartphone companies and/or cloud providers such as Samsung SmartThings, Apple, Amazon Alexa, Google Nest, Xiaomi, and IFTTT etc. [7, 11, 27, 31, 44, 50]. Finally, Infrastructure-centric solutions are ones in which a vendor offers its infrastructure as a platform on which clients can integrate different devices and construct and run IoT applications. All the major cloud providers [13, 14, 26] as well as Tuya Smart [46] offer such platforms.

行业解决方案。当前的物联网（IoT）领域包括多种解决方案，可以大致分为三种形式：设备中心、应用中心和基础设施中心。设备中心的解决方案是指供应商以其自有设备为基础，提供可选的附加软件和云服务。这类供应商通常是设备制造商，例如飞利浦、LIFX、Geeni、BOSE、Dyson 和 WYZE [5, 15, 21, 25, 35, 40, 49]。dSpace 可以集成这些设备中心的解决方案，正如我们在 §6 中所示。我们使用“应用中心”一词来指代那些供应商开发了一个应用程序，并公开提供 API，允许异构设备与该应用或服务进行集成的解决方案。这类供应商通常是智能手机公司和/或云服务提供商，例如三星 SmartThings、苹果、亚马逊 Alexa、谷歌 Nest、小米和 IFTTT 等 [7, 11, 27, 31, 44, 50]。最后，基础设施中心的解决方案是指供应商提供一个平台，客户可以在此平台上集成不同的设备，构建并运行物联网应用程序。所有主要的云服务提供商 [13, 14, 26] 以及涂鸦智能 [46] 都提供此类平台。

IoT form

1. Device-centric: vendor starts with their own device(s) and then offers optional add-on software and cloud services for these devices
   • vendor controls and offers identical service
2. App-centric: vendor has built an application and they expose APIs that allow heterogeneous devices to integrate with this application or service.
   • vendor offers and uses API to extend
3. Infrastructure-centric: vendor offers its infrastructure as a platform on which clients can integrate different devices and construct and run IoT applications
   • platform / cloud

The majority of the above are closed or proprietary solutions. However, there are multiple industry consortia and communities developing open source solutions, including: EdgeX Foundry [23], OM2M [22], KubeEdge [32], Home Assistant [29], and OpenHAB [38]; and they typically offer open source SDKs, device services, communication protocols, pub-subs, UI/UX, and automation modules.

上述大多数解决方案是封闭或专有的。然而，也有多个行业联盟和社区正在开发开源解决方案，包括：EdgeX Foundry [23]、OM2M [22]、KubeEdge [32]、Home Assistant [29] 和 OpenHAB [38]；这些平台通常提供开源 SDK、设备服务、通信协议、发布-订阅（pub-sub）机制、UI/UX 和自动化模块。

Research. There is a large body of prior research in IoT systems closely relevant to dSpace, including: HomeOS [58], BOSS [57], Brick [52], XBOS [60], Beam [72], and Bolt [62]. Among these pioneering works, HomeOS offers PC-like abstractions for programming devices as peripherals and enables cross-device tasks [58]. BOSS provides common system primitives and services such as a query language for metadata, timeseries analytics, and fault-tolerant control for commercial buildings [57]. Beam proposes inference graphs to abstract away sensing and inference tasks, thus simplifying the development of IoT applications [72].

研究。在物联网（IoT）系统领域，有大量与 dSpace 密切相关的先前研究，包括：HomeOS [58]、BOSS [57]、Brick [52]、XBOS [60]、Beam [72] 和 Bolt [62]。在这些开创性工作中，HomeOS 提供了类似 PC 的抽象来编程设备作为外设，并支持跨设备任务的执行 [58]。BOSS 提供了常见的系统原语和服务，如元数据查询语言、时间序列分析和商业建筑的容错控制 [57]。Beam 提出了推理图（inference graph），将传感和推理任务抽象化，从而简化了物联网应用的开发 [72]。

Relation to dSpace. In dSpace, we take inspiration from all of the above and introduce new concepts and abstractions to further simplify the construction of sophisticated smart-space scenarios: first-class and adaptive composition, flexible and higher-level abstractions, intent reconciliation, and support for richer policies that enable runtime adaptation, shared control, delegation, and so forth. As we demonstrate through our system evaluation in §6, the novel features simplify the development of smart space applications.

与 dSpace 的关系。在 dSpace 中，我们从上述所有工作中汲取灵感，并引入了新的概念和抽象，以进一步简化复杂智能空间场景的构建：一流的和自适应的组合、灵活的高层次抽象、意图协调以及支持更丰富的策略，能够实现运行时适配、共享控制、委托等功能。正如我们在 §6 的系统评估中所展示的，这些新颖的特性简化了智能空间应用的开发。

Programming frameworks. As a programming framework, dSpace is inspired by prior research on modular frameworks in other domains - e.g., Malt [70] and Orion [59] for network management, Click [64] for packet-processing, Spark and Hadoop [8, 9] for analytics, Ray and Tensorflow [51] for AI/ML etc. dSpace differs from these in the nature of its abstractions and features which result from the needs and challenges of smart spaces.

编程框架。作为一种编程框架，dSpace 受到了其他领域模块化框架研究的启发 —— 例如，Malt [70] 和 Orion [59] 用于网络管理，Click [64] 用于数据包处理，Spark 和 Hadoop [8, 9] 用于分析，Ray 和 Tensorflow [51] 用于人工智能/机器学习等。与这些框架不同，dSpace 在抽象和特性上有所不同，这些特性源于智能空间的需求和挑战。

Limitation and future works. There are important aspects of IoT system design that we do not cover in this paper. As mentioned in §2.4, current dSpace doesn’t offer UI/UX support (e.g., a GUI) for non-technophile users. We plan to provide such support via integrating or porting existing UI/UX solutions [17, 28]. Besides, we’ve evaluated dSpace under home automation scenarios and plan to extend our evaluation to multi-occupancy homes [78] and to beyond home contexts such as offices, retail locations, and campuses. Finally, we are exploring techniques to provide safety and privacy guarantees in dSpace, e.g., protecting users from unsafe IoT device states via verification and/or enforcing security policies.

局限性与未来工作。在物联网系统设计中，本文未涵盖一些重要方面。如 §2.4 中提到的，目前的 dSpace 不提供非技术用户的 UI/UX 支持（例如 GUI）。我们计划通过集成或移植现有的 UI/UX 解决方案 [17, 28] 来提供此类支持。此外，我们已经在家庭自动化场景下对 dSpace 进行了评估，并计划将评估扩展到多住户家庭 [78]，以及家庭之外的场景，如办公室、零售地点和校园。最后，我们正在探索提供安全性和隐私保证的技术，例如，通过验证和/或执行安全策略来保护用户免受不安全物联网设备状态的影响。

Conclusion

This paper presents dSpace, a framework for building smart space applications. With dSpace, developers can flexibly compose heterogeneous devices into versatile smart spaces that are exposed to users via high-level abstractions which are easier to program and customize. dSpace’s main contribution lies in identifying the design principles and abstractions for the smart-space domain: high-level abstractions via hierarchical composition, decoupled digivice vs. digidata, embedded policies, intent reconciliation, and adaptive composition via policy-driven mount and yield. We show how these abstractions map to the well-known microservices design paradigm and how this paradigm should be adapted to the domain of smart-space applications. We validate our design using real-world IoT devices and scenarios. Compared to existing smart-space systems, dSpace offers: minimalism, better modularity, ease of customization via rich composition and policies, and an open implementation.

本文提出了 dSpace，一个用于构建智能空间应用的框架。通过 dSpace，开发者可以灵活地将异构设备组合成多功能智能空间，并通过更易于编程和定制的高层抽象向用户展示这些空间。dSpace 的主要贡献在于为智能空间领域识别了设计原则和抽象：通过层次化组合实现的高层抽象、解耦的 digi 设备与 digi 数据、嵌入式策略、意图协调，以及通过策略驱动的挂载和暴露实现的自适应组合。我们展示了这些抽象如何映射到著名的微服务设计范式，并说明如何将该范式适应于智能空间应用领域。我们使用现实世界的物联网设备和场景来验证我们的设计。与现有的智能空间系统相比，dSpace 提供了：简约性、更好的模块化、通过丰富的组合和策略实现的定制化简便性，以及一个开放的实现。