Yiwen Shen

Flexfly: enabling a reconfigurable dragonfly through silicon photonics

The Dragonfly topology provides low-diameter connectivity for high-performance computing with all-to-all global links at the inter-group level. Our traffic matrix characterization of various scientific applications shows consistent mismatch between the imbalanced group-to-group traffic and the uniform global bandwidth allocation of Dragonfly. Though adaptive routing has been proposed to utilize bandwidth of non-minimal paths, increased hops and cross-group interference lower efficiency. This work presents a photonic architecture, Flexfly, which “trades” global links among groups using low-radix Silicon photonic switches. With transparent optical switching, Flexfly reconfigures the inter-group topology based on traffic pattern, stealing additional direct bandwidth for communication-intensive group pairs. Simulations with applications such as GTC, Nekbone and LULESH show up to 1.8× speedup over Dragonfly paired with UGAL routing, along with halved hop count and latency for cross-group messages. We built a 32-node Flexfly prototype using a Silicon photonic switch connecting four groups and demonstrated 820 ns interconnect reconfiguration time.

Software-defined networking control plane for seamless integration of multiple silicon photonic switches in Datacom networks

Silicon photonics based switches offer an effective option for the delivery of dynamic bandwidth for future large-scale Datacom systems while maintaining scalable energy efficiency. The integration of a silicon photonics-based optical switching fabric within electronic Datacom architectures requires novel network topologies and arbitration strategies to effectively manage the active elements in the network. We present a scalable software-defined networking control plane to integrate silicon photonic based switches with conventional Ethernet or InfiniBand networks. Our software-defined control plane manages both electronic packet switches and multiple silicon photonic switches for simultaneous packet and circuit switching. We built an experimental Dragonfly network testbed with 16 electronic packet switches and 2 silicon photonic switches to evaluate our control plane. Observed latencies occupied by each step of the switching procedure demonstrate a total of 344 µs control plane latency for data-center and high performance computing platforms.

Flexible Architecture and Autonomous Control Plane for Metro-Scale Geographically Distributed Data Centers

Enterprises and cloud providers are moving away from deployment of large-scale data centers and towards small- to mid-sized data centers because of their lower implementation and maintenance costs. An optical metro network is used to provide connectivity among these data centers. The optical network requires flexibility on bandwidth allocation and various levels of Quality of Service to support the new emerging applications and services including the ones enabled by 5G. As a result, next generation optical metro networks face complex control and management issues that needs to be resolved with automation. We present a converged inter/intra data center network architecture with an autonomous control plane for flexible bandwidth allocation. The architecture supports both single-rate and multi-rate data planes with two types of physical layer connections (Background and Dynamic) that provide connections with strict bandwidth and latency requirements. We demonstrate autonomous bandwidth steering between two data centers on our prototype. Leveraging a simulation platform, we show up to 5

Autonomous network and IT resource management for geographically distributed data centers

\times

Flexible network architecture and provisioning strategy for geographically distributed metro data centers

lower transmission times and 25% less spectrum usage compared with the single-rate conventional non-converged networks. This is a significant improvement in the data center network performance and energy efficiency.

tSDX: Enabling Impairment-Aware All-Optical Inter-Domain Exchange

As enterprises struggle to meet the demand for elastic cloud computing services in distributed data centers (DCs), a significant portion of the total expenditure is spent on the procurement and maintenance of server and network equipment. Network and server resource utilization efficiency is therefore crucial to minimize overall costs and increase return on investment. We present an adaptive and synergistic management of both network and IT resources with an autonomous control plane on a converged inter/ intra-DC network to address this issue. The control plane uses its awareness of both network and IT resource usage to provide dynamic resource allocations to meet quality of service guarantees, while also allowing for reduction in operational expenses by consolidating virtual machines to maintain high CPU usage on the active servers. We built a three-node inter-DC experimental testbed to demonstrate the feasibility of our concept and show that our algorithm successfully provisions both network resources and consolidates IT resources under various load conditions.

Transparent software-defined exchange (tSDX) with real-time OSNR-based impairment-aware wavelength path provisioning across multi-domain optical networks

The fifth generation of mobile networks (5G) is expected to introduce new services with strict end-to-end delay requirements. For this reason, service providers and network operators are increasingly relying on geographically distributed metro data centers (DCs) to bring services closer to end-users and reduce delivery time. The metro DCs frequently exchange data for different purposes, such as backup and load balancing. Some of these data transfers require guaranteed low delays. Meanwhile, efficient use of network resources is necessary to limit the cost of the network infrastructure. To address these issues, in this paper, we propose a converged intra- and inter-DC network architecture and a dynamic provisioning strategy that are able to (i) efficiently support different classes of service, (ii) offer fast data transfers among metro DCs, and (iii) enable efficient utilization of network resources. Simulation results show that the proposed network architecture and provisioning strategy achieve at least two times faster average data transfer between DCs and better network resource utilization compared with conventional solutions. We also present a prototype and an extensive set of experimental results, thus proving the implementation feasibility and effectiveness of the proposed approach.