Data centers are experiencing an exponential increase in the amount of network traffic that they have to sustain due to cloud computing and several emerging web applications. To face this network load, large data centers are required with thousands of servers interconnected with high bandwidth switches. Current data center networks, based on electronic packet switches, consume excessive power to handle the increased communication bandwidth of emerging applications. Optical interconnects have gained attention recently as a promising solution offering high throughput, low latency and reduced energy consumption compared to current networks based on commodity switches. This paper presents a thorough survey on optical interconnects for next generation data center networks. Furthermore, the paper provides a qualitative categorization and comparison of the proposed schemes based on their main features such as connectivity and scalability. Finally, the paper discusses the cost and the power consumption of these schemes that are of primary importance in the future data center networks.

A Survey on Optical Interconnects for Data Centers

High-performance computing systems are of steadily growing interest to provide new levels of computational capability for an increasing range of applications. The growing use of and dependence on optical interconnects to meet these system's scaling bandwidth demands has given rise to “computercom” as a distinct market segment, alongside the traditional datacom and telecom markets. This paper discusses the trends, requirements, tradeoffs, and potential technologies for this market.

Optical Interconnects for High-Performance Computing

Recently, optical orthogonal frequency-division multiplexing technology has attracted intensive research interest because spectrum-sliced elastic optical networks (EONs) can be constructed based on it. In this paper, we investigate how to serve multicast requests over EONs with multicast-capable routing, modulation level, and spectrum assignment (RMSA). Both EON planning with static multicast traffic and EON provisioning with dynamic traffic are studied. For static EON planning, we formulate two integer linear programming (ILP) models, i.e., the joint ILP and the separate ILP. The joint ILP optimizes all multicast requests together, while the separate ILP optimizes one request each time in a sequential way. We also propose a highly efficient heuristic that is based on an adaptive genetic algorithm (GA) with minimum solution revisits. The simulation results indicate that the ILPs and the GA provide more efficient EON planning than the existing multicast-capable RMSA algorithms that use the shortest path tree (SPT) and the minimal spanning tree (MST). The results also show that the GA obtains more efficient EON planning results than the separate ILP with much less running time, as it can optimize all multicast requests together in a highly efficient manner. For the dynamic EON provisioning, we demonstrate that the GA is also applicable, and it achieves lower request blocking probabilities than the benchmark algorithms using SPTand MST.

Efficient resource allocation for all-optical multicasting over spectrum-sliced elastic optical networks

Leveraging the spatial modes of multimode waveguides using mode-division multiplexing on an integrated photonic chip allows unprecedented scaling of bandwidth density for on-chip communication. Switching channels between waveguides is critical for future scalable optical networks, but its implementation in multimode waveguides must address how to simultaneously control modes with vastly different optical properties. Here we present a platform for switching signals between multimode waveguides based on individually processing the spatial mode channels using single-mode elements. Using this wavelength-division multiplexing-compatible platform, we demonstrate a 1×2 multimode switch for a silicon chip that routes four data channels with low (<−16.8  dB) crosstalk. We show bit-error rates below 10−9 and power penalties below 1.4 dB on all channels while routing 10 Gb/s data when each channel is input and routed separately. The switch exhibits an additional power penalty of less than 2.4 dB when all four channels are simultaneously routed. These results enable individual processing of multimode signals and high-bandwidth, flexible optical networks.

On-chip mode-division multiplexing switch

Optical performance monitoring (OPM) is the estimation and acquisition of different physical parameters of transmitted signals and various components of an optical network. OPM functionalities are indispensable in ensuring robust network operation and plays a key role in enabling flexibility and improve overall network efficiency. We review the development of various OPM techniques for direct-detection systems and digital coherent systems and discuss future OPM challenges in flexible and elastic optical networks.

Optical Performance Monitoring: A Review of Current and Future Technologies

A complete review of the data vortex optical packet switched (OPS) interconnection network architecture is presented. The distributed multistage network topology is based on a banyan structure and incorporates a deflection routing scheme ideally suited for implementation with optical components. An implemented 12-port system prototype employs broadband semiconductor optical amplifier switching nodes and is capable of successfully routing multichannel wavelength-division multiplexing packets while maintaining practically error-free signal integrity (BER < 10-12) with median latencies of 110 ns. Packet contentions are resolved without the use of optical buffers via a distributed deflection routing control scheme. The entire payload path in the optical domain exhibits a capacity of nearly 1 Tb/s. Further experimental measurements investigate the OPS interconnection network's flexibility and robustness in terms of optical power dynamic range and network timing. Subsequent experimental investigations support the physical layer scalability of the implemented architecture and serve to substantiate the merits of the data vortex OPS network architectural paradigm. Finally, modified design considerations that aim to increase the network throughput and device-level performance are presented.

The Data Vortex Optical Packet Switched Interconnection Network

Breakthroughs in silicon photonics are changing the economics of network hardware and are enabling new directions in computer networking, compelling us to reexamine our previous assumptions regarding computer and network architectures. One such area that bears further study is the choice of network topology in a world where the distance between nodes is largely irrelevant, and where there is no inherent economic advantage to spatial locality. Network architects should be able to select topologies based on their merits and not merely from tradition. Combined with relaxed system-level packaging constraints, silicon photonics will likely lead to future datacenter architectures that look very different from today’s solutions. We present the design and implementation of Topanga, a packet switch optoASIC with the requisite technologies to free datacenter topologies from constraints induced by limited electrical reach. By means of our platform for integrating silicon photonics inside the package, Topanga provides optical I/O with lower energy per bit at lower cost per port than existing solutions.

Reimagining datacenter topologies with integrated silicon photonics

Future data centers will require novel, scalable memory architectures capable of sustaining high bandwidths while still achieving low memory access latencies. Electronic interconnects cannot meet the challenges presented by the need for multi-terabit off-chip memory data paths. In this work, the electronic bus between main memory and its host processor is replaced with a circuit-switched optical interconnection network. We investigate the impact of our optically connected memory system on large-scale architectures and experimentally validate the protocol using field-programmable gate array based processor nodes and a custom-designed memory controller. The processor communicates all-optically with multiple synchronous dynamic random access memory nodes using 4 × 2.5-Gb/s wavelength-striped payloads, operating error free with bit-error rates less than 10-12.

Building Data Centers With Optically Connected Memory

Wavelength-striped optical packet multicasting comprises a potentially important functionality for future energy-efficient network applications. We report on two multicast-capable architectures to experimentally demonstrate multiwavelength packet multicasting in an optical switching fabric testbed. The first design uses programmable packet-splitter-and-delivery that simultaneously supports the nonblocking unicast, multicast, and broadcast of high-bandwidth optical packets with parallel switches. This realization achieves the error-free multicasting of optical messages with 810 Gb/s payloads, with confirmed bit-error rates less than , and scalability of per-channel data rates to 40 Gb/s. We then introduce a second multistage multicasting architecture with lower hardware and energy costs, with the design trade-off of more complex routing logic; the experimental demonstration shows the successful switching and error-free multicasting of 8 10 Gb/s optical packets. The energy costs in terms of the capital and operational expenditures are then compared for the two designs, showing the benefits of the second multicast architecture.

Broadband Multicasting for Wavelength-Striped Optical Packets

Wavelength-selective spatial routing is proposed for photonic networks and demonstrated on an electro-optic microring switch. This technique yields greater path diversity in photonic networks, enabling improved overall network performance.

Caroline P. Lai

Papers

The Data Vortex Optical Packet Switched Interconnection Network

Reimagining datacenter topologies with integrated silicon photonics

Building Data Centers With Optically Connected Memory

Broadband Multicasting for Wavelength-Striped Optical Packets

Data transmission using wavelength-selective spatial routing for photonic interconnection networks