Houman Rastegarfar

Analysis of large-scale multi-stage all-optical packet switching routers

All-optical packet switching can overcome limitations of electronic switches in terms of power consumption, speed, cost, and footprint. Switch architectures combining wavelength converters and fiber delay lines provide tunable routing and contention resolution when used with an N × N arrayed waveguide grating (AWG), a key passive optical component to bypass electronic processing limitations. An AWG passively routes either single or multiple input port wavelengths to output ports. A single wavelength per port strategy reduces crosstalk within the AWG, but drastically increases the dimensionality of the device. AWG design constraints due to bandwidth limitations and fabrication processes limit the port number for the foreseeable future to under 100. In order to scale optical switches to emerging network requirements, we must use multiple wavelengths per port. In this paper, we examine several optical router architectures for data center applications using multiple wavelengths per port, and quantify the physical layer impairments. We consider not only the AWG crosstalk, but also Q-factor degradation caused by the multiple wavelength conversions occurring when a packet is buffered for contention resolution. We present the results as a function of the number of recirculations for on-off-keying (OOK) signal formats. While previous work has addressed this issue in terms of accumulated loss, we focus on accumulated intensity noise due to crosstalk and amplified spontaneous emission (ASE). We compare the routing performance of each architecture, and we point out that the AWG crosstalk and accumulated ASE noise during packet recirculation are both critical to the routing performance.

Cross-Layer Performance Analysis of Recirculation Buffers for Optical Data Centers

Recirculation buffer modules combining arrayed waveguide gratings (AWGs), tunable wavelength converters (TWCs), and fiber delay lines (FDLs) have been proposed to bypass the existing switching bottlenecks in massive-scale data centers. Performance studies of such subsystems are devoted exclusively to either the network layer or the physical layer aspects. Network layer studies consider packet drops only due to limited buffering capacity and ignore the critical role of the physical layer in degrading signal quality. Purely physical layer studies, on the other hand, are oblivious to contention-based drops and load transients. As a result, neither approach is able to estimate accurately the performance characteristics of buffer modules as key elements in optical data centers. In this theoretical work, we integrate the network layer and the physical layer effects into a single analysis framework to compare various recirculation buffer module designs in terms of throughput, delay, signal quality, and complexity. We primarily compare the designs in terms of two metrics: maximum operating load and Q-factor degradation impact. Our Monte Carlo simulations indicate that Q-factor degradation has the dominant role in determining the buffer module performance over a wide range of load values, resulting in significant bandwidth limitations. In order to implement optical packet switching in data centers, tradeoffs between physical layer quality requirements and forward error correction (FEC) overheads should be carefully investigated.

OOK Q-factor degradation in scalable optical switches

All-optical switching has been proposed to overcome the limitations of electronic switches in terms of scalability, speed, footprint, and power consumption. A key passive optical component to bypass electronic processing limitations is the arrayed waveguide grating (AWG). Switch architectures combining wavelength converters and fiber delay lines provide tunable routing and contention resolution when used with AWGs. An AWG passively routes either single or multiple input port wavelengths to its output ports. A single wavelength per port strategy reduces crosstalk within the AWG, but drastically increases the dimensionality of the device. Physical constraints on AWG design limit the port number for the foreseeable future to under 100. To scale optical switches to emerging network requirements, we can use multiple wavelengths per port. In this paper we examine one multiple wavelength per port architecture and quantify the physical layer impairments due not only to the AWG crosstalk, but also Q-factor degradation due to multiple wavelength conversions, and as a function of the number of recirculations in the contention resolution delay lines. While previous work has addressed this issue in terms of accumulated loss, we focus on accumulated relative intensity noise and amplified spontaneous emission.

A high-performance network architecture for scalable optical datacenters

We propose a novel WDM interconnection scheme for massive-scale datacenters based on arrayed waveguide grating (AWG) devices. We employ optical burst switching and buffer sharing to maximize the datacenter performance gains.

Optical switching and scalability in datacenters

We compare network designs in terms of scale, density, bandwidth, and uniformity of connectivity. We consider all-electronic switching, all-optical switching and hybrid switching with optical WDM interconnection for a datacenter with 10 million microprocessor cores.

Emulation of Optical PIFO Buffers

With recent advances in optical technology, we are closer to building all-optical routers than ever before. A major problem in this area, however, is the lack of all-optical memories similar to what we have in electronics. To overcome this problem, recently, there have been several proposals that show how we can emulate First-In First-Out (FIFO) queues using a combination of fiber delay lines and switches. Unfortunately, FIFO queues cannot be used for implementing many link scheduling policies including weighted fair queuing, weighted round-robin, or strict priority, which are essential components of any modern router today. n nIn this paper, we introduce an architecture based on fiber delay lines and optical switches that can be used for emulating Push-In First-Out (PIFO) queues. In a PIFO queue, an incoming packet can be pushed anywhere in the queue, and therefore it can be used for the implementation of various link scheduling policies. We describe a scheduling algorithm for this architecture and show that with a small speedup, we can build a PIFO queue of size N-1 using only O(log2 N) 3×3 optical switches. The resulting system has a minimum reliability of 99.5%, and even for the small portion of departure requests that cannot be fulfilled immediately, the requested packet is ready to depart within approximately five time slots from the request time.

Cross-layer study of optical burst switches for next-generation datacenters

We analyze the performance of optical burst switches, including physical layer impairments, via the use of Monte Carlo techniques. We compare two designs by investigating the impact of Q-factor degradation on throughput and latency.

WDM Recirculation Buffer-Based Optical Fabric for Scalable Cloud Computing

Optical packet switching is the next-generation disruptive technology for massive cloud data centers. All-optical routers, supporting wavelength parallelism, accelerate the execution of applications and facilitate server virtualization. Despite the recent advances in optical technologies, the greatest challenge in the realization of optical routers is the lack of a mature solution for buffering optical signals. Fiber delay lines (FDLs) have been proposed to emulate electronic buffering by delaying contending optical packets for a fixed amount of time. However, the practical limitation on the number of FDLs in a router requires its ports to be run at low utilization, sacrificing a significant portion of network capacity. In this paper, we introduce a novel modular router architecture based on wavelength-division multiplexed (WDM) recirculation buffers as a solution to the problem of limited buffering capacity. We propose a load-balancing scheduler to maximize throughput. Our mathematical analysis and Monte Carlo simulations show that the consolidation of optical buffers in WDM FDLs accompanied with internal load balancing leads to a virtually lossless router, resilient to data center traffic anomalies. The load balancing further minimizes the number of packet recirculations, leading to negligible queueing latency. Our design supports physical layer scalability and is amenable to large-scale optical integration.

Load balancing in wavelength-routing cloud data centers

We examine the packet loss performance of scalable all-optical routers exploiting fiber delay lines, under non-uniform and bursty data center traffic conditions. Our analysis reveals that a load balancing stage, capable of evenly distributing the incoming traffic in both space and time, is indispensable in wavelength-routing data centers with a distributed-buffer architecture.

Scheduling-Based Load-Balanced Fabric for High-Performance Wavelength-Routing Data Centers

We propose a novel router architecture based on consolidated WDM recirculation fibers as a solution to the challenge of limited optical buffering capacity. Our design employs a distributed load-balancing scheduler that maximizes performance gains.