15 June 2018 in Journal of Optical Communications and Networking
Today’s traditional, hierarchical, full-electronic data centers cannot guarantee quality of service (QoS) for latency-sensitive applications or carrier-grade multitenancy. We propose cloud burst optical-slot switching (CBOSS), an intra-data center network architecture leveraging the original optical components and custom software-defined network (SDN) control to enable traffic engineering. In particular, we demonstrate network slicing on demand with subwavelength granularity and deterministic per-flow QoS guarantees in a 3-top-of-rack (ToR) CBOSS network prototype. We rely on an integrated silicon photonics (SiP) wavelength dropper, a fast-tunable laser, and custom SDN-based controller.
16 May 2019 in Journal of Lightwave Technology
Today's network infrastructure evolves into two seemingly opposite directions: cloudification centralizes functions that used to be distributed for economies of scale, at the expense of latency, while latency-constrained applications are surging. this calls for a new architecture capable of distributed computing: The Edge Cloud network. Future 5G applications will impose strict latency and dynamicity requirements on the Edge Cloud, in intra- and inter-data center networks. The Edge Cloud needs a network infrastructure able to deliver: low latency (~microseconds), deterministic data delivery in time (~nanoseconds jitter) and dynamic reconfiguration (~milliseconds) between objects (antennas, robots) in data centers or across data centers, through a fronthaul network. In this paper we propose, implement and demonstrate Deterministic Dynamic Network (DDN)-based Edge Cloud network. On a real-time testbed we achieve network slicing, low, deterministic latency of only tens of microseconds per-application (per-flow), when competing technologies cannot provide per-flow guarantee. We also show that the network can be dynamically reconfigured at the millisecond timescale.
29 June 2021 in IEEE Journal of Selected Topics in Quantum Electronics
By massively digitalizing production processes, industries expect efficient and secure cooperation of a large number of machines, as well as increasing control and services in edge clouds. Their vision calls for a distributed, converged datacom and telecom infrastructure where performance, can be deterministic per application, i.e. strictly guaranteed, end-to-end, across technologies and across layers. Its success will largely depend on how dynamically it can reuse resources and how cost-effectively it can host mainstream best-effort applications. In this paper, we review the key performance indicators for such an infrastructure while confronting them to thirty use cases reported by industries. We then examine the requirements for the digital infrastructure. We benchmark them against the capabilities of relevant network technologies, while pointing to meaningful enhancements. We particularly discuss Time-Sensitive Networking (TSN), a set of standards extending IEEE 802.1 Ethernet with some deterministic functions, which is considered to be the most promising enabling technology for future industrial networks. We propose to augment its scalability with a novel optical backbone which can interconnect islands of TSN networks while preserving timing.
29 March 2019 in Journal of Optical Communications and Networking
Fifth-generation (5G) systems will introduce several new requirements impacting the mobile transport network. Understanding the impact of 5G systems on transport networks, both wireless and fixed, is essential to the successful introduction of 5G. This paper provides an overview of the ongoing work supporting the specification of open interfaces between the different 5G building blocks. It also presents several technologies that can facilitate effective transport networks, describes the architectural challenges in designing flexible 5G transport networks, and provides dimensioning guidelines.
11 August 2020 in Journal of Optical Communications and Networking
The dynamicity of today’s optical networks is far from its potential. Optical components, such as fast-tunable lasers or semiconductor optical amplifiers, can react on a nanosecond time scale, while the reconfiguration time of optical networks is many orders of magnitude larger, normally above a hundreds of milliseconds timescale. In this work, we address this gap with real-time control plane strategies that enhance the responsiveness of optical networks, specifically in the context of time-critical applications where service determinism is of paramount importance. This context represents an additional challenge since the infrastructure necessary to provide time-wise guarantees increases the complexity of the system under control. We describe in detail the real-time control plane for deterministic and dynamic networks and assess its value through experimental evaluation for the first time to our knowledge of a complete real-time control plane within a multinetwork segment testbed. We prove submillisecond overall reconfiguration time for multinetwork segment environments spanning distances of the order of tens of kilometers.
19 August 2021 in Journal of Optical Communications and Networking
High bandwidth demands combined with low latency applications shifted data processing from the centralized cloud to the edge data center (DC). We propose to leverage the deterministic data delivery in optical deterministic dynamic networks (DDNs) to operate the edge DC in a synchronous manner. We jointly optimize compute and network resources allocation to deliver data just in time to be processed at the compute resources. Numerical results show that our DDN-based approach can (1) accelerate the application total execution time by up to 70% from baseline approaches; (2) reduce performance variance for competing applications, therefore satisfying the service level objective for all; and (3) increase the efficiency of the edge DC in a heterogeneous application environment.