Scaling Low-Latency Optical Packet Switches to a Thousand Ports

Lucente, S. Di; Calabretta, Nicola; Resing, Jacques; Dorren, H.J.S.

doi:10.1364/jocn.4.000a17

Cited by 41 publications

(22 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The figures clearly indicate that the performances worsen increasing the port-count. This behavior is due to the fact that the contention probability increases increasing the port-count as explained in [11]. Simulation results indicate a packet loss lower than 10 −6 with an average latency around 900 ns, and 100% throughput is assured for input traffic up to 0.3 regardless the switch portcount.…”

Section: Numerical Investigationmentioning

confidence: 90%

“…This makes the overall switch configuration time almost independent of the number of ports. As we have already shown in [11], in order to have low configuration time and thus low extra-latency added to the system the employment of a distributed control is necessary. Scaling an OPS based on rearrangeable non-blocking architectures, like Beneš or Banyan, to a large number of ports would require long configuration times due to the need of a centralized controller.…”

Section: Wdm Modular Ops Architecture With Highly Distributed Controlmentioning

confidence: 99%

See 1 more Smart Citation

Numerical and experimental study of a high port-density WDM optical packet switch architecture for data centers

Lucente¹,

Luo²,

Centelles³

et al. 2013

Opt. Express

Self Cite

View full text Add to dashboard Cite

Data centers have to sustain the rapid growth of data traffic due to the increasing demand of bandwidth-hungry internet services. The current intra-data center fat tree topology causes communication bottlenecks in the server interaction process, power-hungry O-E-O conversions that limit the minimum latency and the power efficiency of these systems. In this paper we numerically and experimentally investigate an optical packet switch architecture with modular structure and highly distributed control that allow configuration times in the order of nanoseconds. Numerical results indicate that the candidate architecture scaled over 4000 ports, provides an overall throughput over 50 Tb/s and a packet loss rate below 10 −6 while assuring sub-microsecond latency. We present experimental results that demonstrate the feasibility of a 16x16 optical packet switch based on parallel 1x4 integrated optical cross-connect modules. Error-free operations can be achieved with 4 dB penalty while the overall energy consumption is of 66 pJ/b. Based on those results, we discuss feasibility to scale the architecture to a much larger port count.

show abstract

Section: Numerical Investigationmentioning

confidence: 90%

Section: Wdm Modular Ops Architecture With Highly Distributed Controlmentioning

confidence: 99%

Numerical and experimental study of a high port-density WDM optical packet switch architecture for data centers

Lucente¹,

Luo²,

Centelles³

et al. 2013

Opt. Express

Self Cite

View full text Add to dashboard Cite

show abstract

“…An appropriate scheduling algorithm along with edge electronic buffers enables a contention-free optical burst switching core network. The idea of building large-scale optical burst switches with edge electronic buffers has been the motivation of various research efforts [83][84][85][86].…”

Section: Optical Burst Switching With Edge Electronic Buffersmentioning

confidence: 99%

Optical Switching in Next Generation Data Centers

Testa¹,

Pavesi²

2018

View full text Add to dashboard Cite

This thesis presents advanced optical connectivity solutions for next-generation data centers, spanning millions of processing cores. The wavelength routing of optical packets based on arrayed waveguide gratings (AWGs), tunable wavelength converters (TWCs), and fiber delay lines (FDLs) is a disruptive technology for capacity, footprint, and flexibility requirements of massive cloud data centers. We develop a wavelength-division multiplexed (WDM) design based on Birkhoff-von Neumann architecture as a bitrate transparent, scalable solution to switching bottlenecks in data center networks. As no mature all-optical buffering technology is currently available, central to our design is the buffering strategy that we adopt to store contending packets. Due to the limited number of FDLs, packet drops are common in optical packet switching (OPS) data centers and limit the network throughput. In a practical scenario, physical layer impairments, predominantly amplified spontaneous emission (ASE) noise and crosstalk, degrade Q-factor and lead to additional packet drops. To examine the performance of optical buffer units more accurately, we develop a unified analysis framework, integrating the network layer and the physical layer effects. Using this framework, we refine our architectures and provide guidelines for minimizing the physical layer impact. Besides, optical switch designs should not neglect the diversity of data center traffic patterns when developing connectivity solutions. Empirical studies of data center network traffic reveal that server traffic exhibits strong spatial and temporal correlations. We examine the effectiveness of an AWG-based load balancer with round-robin tuning TWCs in resolving congestion penalties in the data center network. Our cross-layer analysis suggests that the physical layer impairments due to the load balancer hardware could counteract its potential gains and lead to significant throughput degradation. To resolve this challenge, we develop a novel modular router architecture based on WDM shared recirculation buffers that integrates the functions of load balancing and routing and maximizes the optical buffering gains. The router employs a load-balancing scheduler to maximize throughput and minimize delay. Our mathematical analysis and Monte Carlo simulations show that the consolidation of optical buffer slots into WDM FDLs accompanied with internal load balancing leads to a virtually lossless router, resilient to data center traffic anomalies.ii

show abstract

“…This is investigated for a link with flow-control. In [11][12][13], this phenomena is investigated using a variety of analytical and numerical approaches. In brief, it is used that an optical link with a switching system as in Fig Using the approaches, we were able to compute the latency and throughput of the switch as a function of the number of ports (Fig.…”

Section: Architecturesmentioning

confidence: 99%

“…3, only the switch was replaced. It is visible that sub-microsecond latency at high system loads is achievable [11]. The switch architecture is as in Fig.…”

Section: Architecturesmentioning

confidence: 99%