Deflection Routing In Networks Research Articles

An NoC Simulator That Supports Deflection Routing, GPU/CPU Integration, and Co-Simulation

We present deflection routing network on chip simulator (DNOC), a network-on-chip simulator. DNOC is primarily a deflection routing simulator, it simulates custom network topologies with detailed deflection router models, and a basic virtual channel router. DNOC can generate various statistics, such as network latency and power. We evaluate the simulator in three typical use cases. In stand-alone simulation, synthetic traffic generators are used to offer load to the network. In synchronous co-simulation, the simulator is integrated as a module within a larger system simulator with synchronization every simulated cycle. In the faster model-based co-simulation mode, a latency model is built, and retuned periodically at longer time intervals. We demonstrate co-simulation by running applications from the Rodinia and SPLASH-2 benchmark sets on mesh variants. DNOC is also able to run on multicore processors, speeding up the simulation of large networks.

In-channel misrouting suppression technique for deflection-routed networks on chip

Deflection routing, where port-contentions in routers are resolved by intentionally misrouting some of packets along unwanted directions instead of storing them, has been proposed as a promising approach for improving power and area efficiency of large-scale networks on chip (NoCs). However, at high network load, when packets are misrouted more frequently, the cost and energy benefits of this simple routing scheme are offset by the performance degradation. To address this problem, we propose a technique that uses small in-channel buffers to capture some of deflected packets before they take a misrouting hop. The captured packets are then looped-back to the routers where they suffered deflection and routed again. To improve the efficiency of this in-channel misrouting suppression scheme we also slightly modify the routing function of the deflection router by restricting the choice of productive directions for misrouted packets. Evaluations on synthetic traffic patterns show that the proposed misrouting suppression mechanism yields an improvement of 36.2% in network saturation throughput when implemented into the conventional deflection-routed network.

Zero-load predictive model for performance analysis in deflection routing NoCs

We study a static model for 2-D and 3-D networks that accurately represents the average distance travelled by packets under deflection routing, which is a specific form of adaptive routing. The model captures static properties of the network topology and the spatial distribution of traffic, but does not take into account traffic loading and congestion. Even though this static model cannot accurately predict packet latency under high load, we contend that it is a perfect predictor of deflection routing networks’ relative performance under any load condition below saturation, and thus always correctly predicts the optimum network configuration. This is verified through cycle-accurate simulations of congested and uncongested networks with fully adaptive, deflection routing for regular traffic patterns such as uniform random, localised, bursty, and others, as well as irregular patterns in both regular and irregular networks. As the networks with minimal average distance perform best even under high traffic load, the average distance model establishes a robust relation between a static network property, average distance, and network performance under load, providing new insight into network behaviour and an opportunity to identify the optimal network configuration without time-consuming simulations.

Performance Model of Multichannel Deflection-Routed All-Optical Networks With Packet Injection Control

Deflection routing is a feasible approach to resolve the output contention problem in packet-switched networks when buffering of packets is not practical. In this paper, we investigate the performance of multichannel deflection-routed networks with no packet injection control, strict packet injection control, and a simple token-bucket-based packet injection control. The analytical performance models of multichannel deflection-routed networks with strict packet injection control are derived. Simulation results show that the analytical models can accurately predict the performance regardless of the network topology, number of channels, and packet injection control methods. We observed that the end-to-end throughput-delay and the packet re-transmission performance at sources can be largely improved by using simple packet injection control mechanisms such as the proposed token-bucket-based method.

A fully implemented 12 /spl times/ 12 data vortex optical packet switching interconnection network

A fully functional optical packet switching (OPS) interconnection network based on the data vortex architecture is presented. The photonic switching fabric uniquely capitalizes on the enormous bandwidth advantage of wavelength division multiplexing (WDM) wavelength parallelism while delivering minimal packet transit latency. Utilizing semiconductor optical amplifier (SOA)-based switching nodes and conventional fiber-optic technology, the 12-port system exhibits a capacity of nearly 1 Tb/s. Optical packets containing an eight-wavelength WDM payload with 10 Gb/s per wavelength are routed successfully to all 12 ports while maintaining a bit error rate (BER) of 10/sup -12/ or better. Median port-to-port latencies of 110 ns are achieved with a distributed deflection routing network that resolves packet contention on-the-fly without the use of optical buffers and maintains the entire payload path in the optical domain.

Performance Evaluation of a Buffered ShuffleNet with Wavelength Conversion

A novel hybrid contention resolution utilizing optical buffering and wavelength conversion is proposed for deflection routing networks. The corresponding network performance is investigated systematically by means of theoretical analysis and network simulation. Taking two-connected ShuffleNet as an example, we show that a 64-node ShuffleNet indeed performs very well with only single optical buffer and four wavelengths, and can achieve performance close to the store-and-forward case using only single optical buffer and ten wavelengths.

A survey of data multicast techniques, architectures, and algorithms

This article presents a survey of architectures, techniques, and algorithms for multicasting data in communication switching networks. We start with a broadcast architecture using a separate copy network and a routing network. A few versions of this idea using Delta and Benes networks exist. Another multicast architecture is a recycling network where internal nodes act as relay points, accept packets from the switching fabric, and recycle them back into the fabric after relabeling the packets. Next, we give an overview of a system that uses the Boolean splitting multicast algorithm. In this system a nonblocking self routing broadcast banyan copy network has been proposed. The network consists of several components including a running adder network to generate running sums of copy numbers specified in the headers of input packets. We then describe a multicasting technique presented for a different class of switching networks called deflection-routing networks. Finally, the idea of extending a nonblocking network to a three-dimensional structure consisting of multiple parallel planes is also presented. At the end of this article, we compare the efficiencies of the presented multicast architectures.

Evaluation of an ATM LAN constructed with a cyclic deflection-routing network

This paper presents the design and performance evaluation of an ATM local-area network constructed with the Manhattan-street network (MSN). The MSN has a cyclic structure and belongs to the class of deflection-routing networks. The structure of the switch element with the use of shared buffering and flow control mechanism is presented. With shared buffers the network operates efficiently since the occurrence of deflections is minimized. Cell multicasting under deflection routing is handled using the cell recycling technique and through relay and copy nodes. An analytical queueing model is developed for the network and the numerical results of the traffic performance analysis is compared with the results of a separate simulation. It is determined what speed-up factor (speed advantage) the network requires to become a practical LAN. 1 In this paper, speed-up factor and speed advantage are interchangeably used to refer to the internal link to external link speed ratio. 1

Analytical evaluation of improved access techniques in deflection routing networks

This paper presents an extension of a known analytical model for the performance evaluation of nonpriority deflection routing networks in uniform traffic. The extension allows the analysis of improved access techniques. The key features of the analytical technique are described by casting it in a very simple setting: nonpriority hot-potato in a two-connected slotted shufflenet (SN) network. Results are presented for three access techniques: transmit-no-hold (TXNH), transmit-hold (TXH), and bypass queuing (BQ).

A performance model of deflection routing in multibuffer networks with nonuniform traffic

Deflection routing can be used in networks whose stations have the same number of input and output links. Fixed length packets arrive synchronously on the station's input links at the beginning of time slots, and each packet is routed via the output link that offers the shortest path to its destination. Since the number of packet buffers at each output link is finite, the simultaneous contention of two packets for the last buffer of a common output link must be resolved by "deflecting" one of the packets to another output link. Thus, the deflection of a packet could result in the packet following a route that is not a shortest path. The potentially unbounded number of routes that a given packet can take makes analyzing the performance of such networks difficult. In particular, there are no analytical models that can analyze multibuffer deflection-routing networks with nonuniform traffic. Using independence assumptions, the authors develop a performance model of deflection routing that allows to estimate accurately and efficiently the mean transport time and throughput in a network that has any given two-connected topology, multiple buffers at each output port, and an arbitrary traffic matrix.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

A versatile model for predicting the performance of deflection-routing networks

Deflection routing can be used in networks whose stations have the same number of input and output links. Fixed-length packets arrive synchronously on the station's input links at the beginning output link that offers the shortest path to its destination. Since the number of packet buffers at each output link is finite, the simultaneous contention of two packets for the last buffer of the common output link must be resolved by “deflecting” one of the packets according to a specified criterion (e.g. at random, by destination proximity, or by packet age). Deflection routing can therefore be used with as few as one packet buffer per output link. The potentially unbounded number of routes that a given packet can take makes analyzing the performance of such networks difficult. Using independence assumptions, we develop an efficient, high-fidelity performance model of deflection routing that allows us to estimated the mean end-to-end packet delay in a network that has any given two-connected topology, a single packet buffer at each output port, and an arbitrary traffic matrix.

Analytic node model for deflection routing networks

An analytic model is introduced for calculating mean nodal queuing delays in deflection routed networks. In the model formulation, we obtain a mathematically tractable result by assuming that transit traffic arriving to the node in question possesses both spatial and temporal independence. The proposed model is applied to the SIGnet metropolitan area network. The results indicate that in typical subnetworks, reasonable accuracy may be obtained over a wide range of parameter values using link independence assumptions.