DNA: A Dependable NIC Architecture

Objective

The DNA project aims to develop and deploy trustworthy distributed systems on large-scale heterogeneous datacenters (HDCs) such that their dependability properties, i.e., safety, security, and fault-tolerance, can be enforced by the foundational layers of the network hardware stack. Distributed systems have become ubiquitous in modern cloud environments, offering high availability, perfor- mance, and resource scalability. However, ensuring the dependability and performance of distributed systems in large-scale HDCs, which are vulnerable to malicious threats and arbitrary (Byzantine) failures, is a complex task. Existing systems are limited by (a) programmability: They lack a unified, trusted foundation for security and programmability, (b) security: They have vulnerabilities and may be unable to handle malicious attacks and (c) high-performance: They can be slow, especially in complex distributed systems. These limitations force programmers to use ad-hoc methods, which can be expensive and unreliable. The DNA project seeks to resolve the design trade-off between security and performance by introducing a new trustworthy network-level abstraction. DNA realizes this abstraction by proposing a minimalistic, high-performance, and verifiable silicon-root-of-trust as the foundational hardware building block for implementing trustworthy distributed systems. Our implementation of DNA on commodity SmartNICs will revolutionize the way distributed systems are built in cloud environments. By providing a synergistic co-design of hardware-software networking substrate, we abstract out the hardware heterogeneity in cloud environments while providing a minimalistic interface for resource management, isolation, communication, and trust establishment. Overall, we aim to empower programmers and significantly strengthen the dependability properties of widely adopted distributed systems.