On the Feasibility of Optical Circuit Switching for High Performance Computing Systems (original) (raw)
Related papers
Optical interconnection networks for high-performance computing systems
Reports on Progress in Physics, 2012
Enabled by silicon photonic technology, optical interconnection networks have the potential to be a key disruptive technology in computing and communication industries. The enduring pursuit of performance gains in computing, combined with stringent power constraints, has fostered the ever-growing computational parallelism associated with chip multiprocessors, memory systems, high-performance computing systems and data centers. Sustaining these parallelism growths introduces unique challenges for on-and off-chip communications, shifting the focus toward novel and fundamentally different communication approaches. Chip-scale photonic interconnection networks, enabled by high-performance silicon photonic devices, offer unprecedented bandwidth scalability with reduced power consumption. We demonstrate that the silicon photonic platforms have already produced all the high-performance photonic devices required to realize these types of networks. Through extensive empirical characterization in much of our work, we demonstrate such feasibility of waveguides, modulators, switches and photodetectors. We also demonstrate systems that simultaneously combine many functionalities to achieve more complex building blocks. We propose novel silicon photonic devices, subsystems, network topologies and architectures to enable unprecedented performance of these photonic interconnection networks. Furthermore, the advantages of photonic interconnection networks extend far beyond the chip, offering advanced communication environments for memory systems, high-performance computing systems, and data centers.
Optical interconnections within modern high-performance computing systems
Proceedings of The IEEE, 2000
Optical technologies are ubiquitous in telecommunications networks and systems, providing multiple wavelength channels of transport at 2.5-10 Gbps data rates over single fiber-optic cables. Market pressures continue to drive the number of wavelength channels per fiber and the data rate per channel. This trend will continue for many years to come as e-commerce grows and enterprises demand higher and reliable bandwidth over long distances. E-commerce, in turn, is driving the growth curves for single-processor and multiprocessor performance in data-base transaction and Web-based servers. Ironically, the insatiable taste for enterprise network bandwidth, which has driven up the volume and pushed down the price of optical components for telecommunications, is simultaneously stressing computer system bandwidth-increasing the need for new interconnection schemes-and providing for the first time commercial opportunities for optical components in computer systems. This paper will center primarily on the use of optical interconnects within commercial digital computing systems, particularly workstations and servers, and will address mainly board-board interconnects within a single cabinet or box. We feel this is the most likely utilization of optics in commercial computer systems for the next decade. We will also provide a practical analysis of inter-and intrachip optical interconnects and the difficulties they face in real systems.
Lecture Notes in Computer Science, 2013
In response to the need for faster and fatter networks for large-scale HPC cluster systems, hybrid optical/electrical networks have been proposed as an affordable and high-capacity solution. Still, there is no prior work evaluating the performance of HPC workloads over such types of networks. To fill this gap, this work presents a hybrid network architecture comprising commodity-only equipment, shows its price competitiveness against fat-tree alternatives and presents a prototype implementation. We evaluated several HPC workloads over our prototype, showing that our hybrid optical/electrical network manages to significantly accelerate tested workloads, without incurring any extra cost compared to an all-electronic fat-tree network.
Supporting Highly Parallel Computing With a High Bandwidth Optical Interconnect
… OF LEEDS SCHOOL …, 2001
The list of applications requiring high performance computing resources is constantly growing. The cost of interprocessor communication is critical in determining the performance of massively parallel computing systems for many of these applications. This paper considers the feasibility of a commodity processor-based system which uses a free-space optical interconnect. A novel architecture, based on this technology, is presented. Analytical and simulation results based on an implementation of BSP (Bulk Synchronous Parallelism) are presented, indicating than a significant performance enhancement, over architectures using conventional interconnect technology, is possible.
Energy-Efficient Design of a Scalable Optical Multiplane Interconnection Architecture
IEEE Journal of Selected Topics in Quantum Electronics, 2000
As the power dissipation of data centers challenges their scalability, architectures for interconnecting computers, or servers must simultaneously achieve high throughput at peak utilization and power consumption proportional to utilization levels. To achieve this goal, this paper proposes the use of an optical multiplane interconnection network, named space-wavelength (SW) switched architecture, able to route and switch packets between servers (on cards) and between processors within a card (or card ports). SW architecture exploits the space domain to address the destination card and the wavelength domain to address the destination port on a per-packet basis. Scalability and energy efficiency of the considered architecture are quantified and compared to typical single-plane architectures. Not only can the SW multiplane architecture achieve higher throughput by exploiting two switching domains, but its performance is shown to be highly scalable with network utilization. More importantly, higher performance is reached with an energy efficiency superior to single-plane architectures. The excellent energy efficiency is achieved using optical devices with low idle power.
Optical-packet-switched interconnect for supercomputer applications [Invited]
Journal of Optical Networking, 2004
We describe a low-latency, high-throughput scalable optical interconnect switch for high-performance computer systems that features a broadcast-and-select architecture based on wavelength-and space-division multiplexing. Its electronic control architecture is optimized for low latency and high use. Our demonstration system will support 64 nodes with a line rate of 40 Gbit/s per node and operate on fixed-length packets with a duration of 51.2 ns using burst-mode receivers. We address the key system-level requirements and challenges for such applications.
Large data center interconnects employing hybrid optical switching
Proceedings of the 2013 18th European Conference on Network and Optical Communications & 2013 8th Conference on Optical Cabling and Infrastructure (NOC-OC&I), 2013
Current data centers (DCs) networks rely on electronic switching and point-to-point interconnects. When considering future DC requirements, point-to-point interconnects will lead to poor network scalability and large power consumption. For this reason several optical switched interconnects for DCs have been recently proposed. However, the proposed optical switching solutions suffer from low flexibility and are not able to provide service differentiation. Furthermore, very few studies evaluate possible improvements in energy efficiency offered by optical switching solutions. In this paper we introduce a novel architecture of interconnects for DCs based on hybrid optical switching (HOS). HOS combines three different optical switching paradigms, namely circuit, burst and packet switching within the same network. Furthermore, HOS envisages the use a two parallel optical switches, a slow and low power consuming switch for the transmission of data using circuits and long bursts, and a fast switch for the transmission of packets and short bursts. The possibility of choosing between circuits, bursts and packets ensures the flexibility required by future DCs. At the same time, the option to select the most suitable switch technology for each data flow guarantees high transmission efficiency and low power consumption.
Optical interconnects: out of the box forever
IEEE Journal of Selected Topics in Quantum Electronics, 2003
Based on a variety of optimization criteria, recent research has suggested that optical interconnects are a viable alternative to electrical interconnects for board-to-board, chip-to-chip, and on-chip applications. However, the design of modern high-performance computing systems must account for a variety of performance scaling factors that are not included in these analyses.