Theme 1. ARENA Integrator

The Augmented Reality Edge Networking Architecture (ARENA) is the center’s main integration platform. Theme 1 focuses on prototyping compelling applications that disrupt the prevalent “cloud, middle, edge” system design paradigm. Demonstrator applications (autonomous systems, mixed reality, and smart cities) are explored under the unifying “Connected Environments” task. The emphasis of Theme 1 will be on the system infrastructure needed for real-time network observability, visualization, and in-network compute migration and adaptation.

1.1 Composable, Secure Runtime

This task is aimed at pulling together the system elements and building composable runtime with predictability and security properties. This work will be developed in coordination with efforts in other Themes, such as 3 (security) and 6 (programming and resource management). We expect this task to develop the software artifact used to link (connect various I/O components) and launch ARENA programs.

1.2 Connected Environments

The CONIX interpretation of the Spatial Web goes beyond mobile headset/phone platforms in order to support richer human-to-machine as well as machine-to-machine interactions (digital twin runtime platform). One important aspect is to enable tight cooperation between users and other computing systems in this environment. This task now houses our application drivers(autonomous systems, mixed reality, and smart cities). As an application driver task, this draws heavily from all of the Themes.

1.3 Distributed Video Analytics

This task will develop data-flow graph descriptions for distributed visual analytics applications. Nodes in these graphs represent operators which process visual data streams and can be independently optimized, accelerated, or partitioned across network nodes.

1.4 Distributed RF Spectrum Sensing

The main research outcomes for this task are to: (i) characterize network joining, churning, and updating state and its associated overhead; (ii) demonstrating that a distributed architecture can implement the functionality of a centralized one with a constrained performance and power penalty, and integrate with CONIX platforms and ARENA architecture.

1.5 Android Team Awareness Kit (ATAK) XR

We will develop two main technical components: (1) design and integration of a rapidly deployable GPS‐denied localization capability for mobile phone platforms (i.e. ATAK) and (2) integration of a runtime to view and interact with 3D content. The system should remain previous platforms and support upcoming spatial web frameworks.

RELATED PUBLICATIONS

Theme 2. Hardware/Software Platforms

Theme 2 will create the hardware support for mixed reality applications, networking, and reliable and long-lifetime sensor platforms. In Task 2.1, the team will develop highly-constrained energy harvesting devices, whereas Task 2.2 is dedicated to mobile and wearable platforms with advanced communication and sensing capabilities and platforms (such as gateways) to support the challenges of deploying reliable and long lifetime sensors.

2.1 Near-Zero Power Platforms (Dutta, Gupta, Lucia)

The result of this task will be hardware platforms and design methodologies for extremely constrained M-class processor systems with the goal of perpetual energy-harvesting operation. These platforms will expose interfaces that facilitate mapping from high-level application constraints and behavioral policies to hardware platforms that carry out a computation, under extreme environmental constraints, such as energy intermittency.

2.2 Sensing and Interaction (Gupta, Harrison, Hoe, Rabaey, Rowe)

This task will focus on the class of algorithms and hardware platforms for next-generation sensing systems for application-class processors. This includes several interaction technologies, such as wearable speech processing systems, human and scene capture technologies, and mobile interface platforms like AR/VR headsets, drones and other wearables.

2.3 Low Power Intermittent Computing Architectures for ChipSats (Lucia)

The task will develop a vector-dataflow microarchitecture in a RISC-V architecture with vector extensions in an open-source core design for secure, efficient, low-power operation. The task will study intermittence-safe memory models with automatic cache control informed by dynamic dataflow monitored at runtime.

RELATED PUBLICATIONS

Theme 3. Security

The overarching goal of Theme 3 is to develop the foundations for enabling secure and privacy-preserving technologies for future CONIX-enabled applications. To this end, Theme 3 tries to tackle fundamental challenges that encompass policy abstractions for expressing these capabilities, platforms for enforcing these policies, and capabilities for learning to tackle advanced adversarial strategies.

3.1 Trustworthy Components (Parno, Sekar, Srivastava, Stefan)

The focus of this task will be to develop the tools and methodologies CONIX needs in order to construct a secure distributed foundation that comes with formal end-to-end guarantees of correctness, security, and performance.

3.2 Resilient and Secure Networks (Culler, Parno, Sekar, Stefan)

This task will develop the platforms and algorithms CONIX needs in order to construct resilient network security capabilities to handle a continuously evolving attack landscape.

3.3 Secure Programming (Parno, Srivastava, Stefan)

In this task, CONIX will develop the tools and mechanisms that will allow developers to build and safely deploy secure, privacy-preserving applications atop the CONIX platform.

3.4 Enabling Memory-Intensive, In-Network Computing Applications with Programmable Switches and Intelligent Memory (Hoe, Sekar)

This task will develop architectural candidates for integrating programmable switches with in-/near-data processing, demonstrating them with several NFs, including high-fidelity network measurement and monitoring, network telemetry for defense against distributed denial of service (DDoS), and intrusion prevention.

RELATED PUBLICATIONS

Theme 4. Machine Learning

The primary goal of this Theme is to add intelligence to the network and to integrate machine learning, AI, and intelligent control capabilities, throughout the nodes of a network and throughout the various other Themes in the center. This theme is divided into two taks, where one is dedicated to develop and apply machine learning techniques and the other is focused on foundations that will lead to the next generation of ML systems.

4.1 Learning-Enabled Systems (Li, Rabaey, Srivastava)

Our approach is to refine existing and develop novel and specialized machine learning approaches to provide smart algorithms throughout the CONIX infrastructure and that can be applied in many Themes. For example, in-network distributed summarization (see below) will allow massive amounts of information to flow throughout a network without saturating capacities. DDoS attacks will be met with distributed ”defense of service” procedures, where smart actors perpetually monitor network state and behave collectively by acting locally. Our philosophy is that the network is the brain — this means that global behavioral changes in a network can be achieved through local integrated intelligence that is distributed everywhere.

4.2 ML Foundations (Bilmes, Smith)

Problems involving massive unstructured multi-modal streaming real-time data are best served using modern machine learning (ML) and artificial intelligence (AI) approaches, where humans code indirectly (i.e., write algorithms to learn other algorithms) rather than directly. Our contribution will develop ML and AI foundations within the CONIX networked and distributed computing substrate. This work will include stream processing, data summarization and federated machine learning.

4.3 Learning for Security and Privacy (Bilmes, Sekar)

The focus of this task will be to develop the tools, algorithms, and methodologies to model advanced security and privacy threats and adaptive adversaries in a CONIX deployment.

RELATED PUBLICATIONS

Theme 5. Communication, Positioning and Control

This theme will address the necessary wireless communication, localization and timing primitives that are necessary for the correct and efficient operation of distributed perception-cognition-action applications.

5.1 Positioning, Navigation and Timing (Dutta, Govindan, Rowe, Srivastava)

The Task will develop building blocks for providing applications with access to spatial and temporal information and associated services, particularly those needed by the driver applications in Theme 1. This includes multiple indoor localization systems ranging from visible light communication and ultrasound to Ultra-Wide Band ranging radios and RF beamforming approaches.

5.2 Wireless Communication (Cabric, Dutta, Rowe, Wawrzynek)

This task will address some of the communication building blocks for wireless platforms. We plan to expose low-level radio functionality to the software stack and develop a software-defined radio infrastructure for low-power wide-area networking. This will enable better network resource usage and increased capabilities. This task will also work closely with ComSenTer in order to identify system solutions for future high-bandwidth communication channels including localization, hand-off and networks where beamforming and antenna arrays can be managed by autonomous agents.

5.3 Networked Control (Tabuada, Tomlin)

This task develops the scientific principles and the design methodologies for networked control over distributed computing substrates, addressing control and physical state estimation as a network service, providing seamless transition between local computation and global properties, with an emphasis on safety, and safe learning.

5.4 Panoptes: Real-Time 3D Modeling of Large Urban Spaces (Govindan, Srivastava, Rowe)

This task is developing 3D sensors for increased situational awareness. This work has the potential to enhance our ARENA demonstrator significantly and pushes the frontier of edge computing and wireless networking.

RELATED PUBLICATIONS

Theme 6. Programming and Resource Management

The CONIX context presents a unique challenge in the development of robust distributed systems with a high degree of flexibility and programmability, extreme heterogeneity of components, and coordination mechanisms. Theme 6 will tackle this challenge.

6.1 Programmer-visible Abstractions (Lucia, Bodik, Gupta, Srivastava, Stefan, Govindan, Dutta)

This task defines macroprogramming abstractions for single-tier programming of the distributed system at scale, spanning from sensors on the edge to compute servers in the core. The abstractions will include new hybrid information flow mechanisms to ensure security (see Task 3.2).

6.2 Coordination and Mapping (Srivastava, Dutta, Bodik, Govindan)

This task will define and implement the system support to map programs written in our novel application-level abstractions onto the abstractions of the target virtual platforms that support key CONIX applications, translating a specification into functionality. This task is closely related to Task 6.1, which permits a logically centralized specification of application logic and correctness checking; our task permits high-level specification of performance goals and optimization hints. Today’s cloud systems have pioneered similar approaches but cannot be easily extended to a multi-tier distributed computing substrate with heterogeneity in computing and communication capabilities across tiers.

6.3 Programming for Reconfigurable Hardware (Hoe, Wawrzynek)

This task will perform research on the support for platform-level heterogeneous hardware configurations that include single-chip SOCs with Field Programmable Gate Arrays (FPGA) fabrics. These devices can be used as an optimization option that trades portability for performance and efficiency.

6.4 Platform Support for a Composable, Secure Common Runtime (Parmer)

Task 6.4 will be developed in connection with Theme 1, and produce research towards a runtime with strong predictability and security properties to build distributed CONIX applications. This will include compiler and runtimes specialized for small embedded devices and runtime components to support density, elasticity, and controlled latency.