Prioritized Shufflenet Routing in TOAD based 2X2 OTDM Router.

Size: px

Start display at page:

Download "Prioritized Shufflenet Routing in TOAD based 2X2 OTDM Router."

Sophia Watson
5 years ago
Views:

1 Prioritized Shufflenet Routing in TOAD based 2X2 OTDM Router. Tekiner Firat, Ghassemlooy Zabih, Thompson Mark, Alkhayatt Samir Optical Communications Research Group, School of Engineering, Sheffield Hallam University, Pond Street, Sheffield, S1 1WB. U.K. Tel: , Fax: , Abstract- Shufflenet [1, 2] is a multihop packet-switched network topology, which has been widely used in the optical communication. The adaptive routing algorithm used in shufflenet provides multiple shortest paths from source to destination. Here shufflenet is applied for optical time division multiplexing (OTDM) based networks, where nodes are composed of a terahertz optical asymmetric demultiplexer (TOAD) based 2x2 routers. Here we have introduced a simple priority scheme, where packets are routed via the transmission port (having lower crosstalk compared with the reflection port) of the router as much as possible. Simulation results show that transmission-to-reflection ratio (TRR) greater than one is achievable over a wide range of system load. Keywords- Shufflenet, Lightwave Network, TOAD, OTDM, Interconnection Networks, Routing, Multihop Network. 1. Introduction The growing demand for increased network capacity has generated interest in the development of ultra fast all optical communication networks, which overcomes the bottleneck of opto-electrical (O/E) and electro-optical (E/O) conversions. Wavelength division multiplexing (WDM) has attracted most of the researchers attention. However, recent work reported on OTDM based network have shown that it can achieve higher transmission speeds, and provide higher scalability, throughput, higher latency and reliability [3]. There are number of network topologies that have been applied to multihop OTDM networks. Here we consider one such a scheme known as shufflenet, which is a multihop topology. In multihop packetrouting networks, packets traverse a number of intermediate nodes until they arrive at their destination. At each node, switching (routing) decisions are made by simply processing only a small fraction of the information contained within the packet header. The routing algorithm used at each node is identical, that decides on the best possible port allocation to the incoming packets. This decision is given independently, without regard for the routing decisions made at the rest of the nodes within the network. Performance analysis of the classical shufflenets has been thoroughly investigated in previous studies [4, 5]. This paper introduces a priority scheme in order to improve the performance of the shufflenet using a TOAD based 2x2 router at each node for OTDM based packet routing network. Figure 1 shows the logical node structure of a TOAD based 2x2 router [3]. Previous studies have shown that in TOAD based routers the crosstalk observed at the output port is not the same. The crosstalk is slightly higher at the reflection port [3]. Therefore, the implication of this in a network environment would be to select routes, which avoids the reflection output ports as much as possible in order to minimise the overall crosstalk observed by a packet. In this paper, a priority scheme for shufflenet

2 has been developed, which minimises the usage of the reflection port and maximises the usage of the transmission port. This selection of output port does not in any way effect the performance (such as through put, average hop distance etc.) of the routing algorithm. The rest of this paper is organized as follows: In Section 2, structural properties of the shufflenet is described, static and adaptive routing algorithms for shufflenet are given. Transmission/Reflection ratio is defined and possible worst and best case path analysis are given in Section 3. Section 4 shows the simulation results for deflection and store-and-forward routing. Results of this paper are summarised in Section 5 with some remarks. A (p,k) Shufflenet consists of N = kp k (k = 1,2,3, and p = 1,2,3) nodes arranged in k columns of p k nodes each and with the last (k-th) and first columns connected together cylindrically where p is in and out degree of the nodes. Figure 2 shows a 24 nodes Shufflenet configuration. In general, node (row, column) is connected to nodes (p.row mod p k, (column + 1) mod k), (p.row mod p k + 1, (column + 1) mod k), (p.row mod p k + p, (column + 1) mod k. Input 0,1 Output 1,1 2,1 0,2 1,2 2,2 3,1 3,2 4,1 4,2 5,1 5,2 6,1 6,2 7,1 7,2 Figure 1. Logical structure of 2X2 TOAD router 2. Shufflenet Shufflenet is classified as a regular multihop network topology, which is a symmetric and homogeneous, i.e., the network looks the same from any node [5]. Thus, the number of hopes from a source to destination is minimised. Due to its cylindrical connectivity pattern (see Figure 2) routing is reliable and fault tolerant. Also, deflection routing can be applied easily with promising results. Figure Node (p=2, k=3) shufflenet composed of 2X2 TOAD router Routing decisions made at each node, are independent of the packet s original source. Therefore, the routing decision is distributed. This also enables simple and easy routing rules to be used. In simple static shufflenet routing, there is only one path from source to destination, which is fixed [4, 6]. However, if a packet can't reach its destination in one pass (i.e it is k-hops away) then it will take one of the many minimum hop routing paths available. A Simple adaptive algorithm exploits this

3 multiplicity of potential routing paths, by routing packets via less-congested paths whenever possible [6], see Figure 2. Generally there are two kinds of packets defined in the adaptive routing. TYPE-S packets that have a single minimum hop path from its current location to destination and TYPE-M packets that have multiple minimum hop paths from its current location to destination. TYPE-M packets have no preference on which output port it goes to. 0,1 1,1 2,1 3,1 4,1 0,2 1,2 2,2 3,2 4,2 3. Transmission/Reflection Ratio (TRR) As discussed previously each node is a 2x2 TOAD router (Figure 1) with has two output ports, with the reflection port (Output-B) having higher crosstalk than the transmission port (Output-A). As shown in Figure 1 packets are inserted (added) to the network when an output port is available and on reaching the final destination they are dropped. In deflection routing if two packets are destined to the same output port, then one of them is deflected. Whereas, in store-and-forward routing if one-level buffer is full, then the packet is deflected to the available port. As shown in Figure 3, packet taking the worst path (all reflection ports) will experience more crosstalk than those taking the best path. The best case scenario is to select a path that links all the transmission ports until the packet reaches its destination. To achieve this we have introduced a priority scheme, where packets going through the transmission port have higher priority over the reflection port. A packet received by a node can be routed arbitrarily to one of the two output ports without affecting the path length to the packet destination. However, when the packet is TYPE-M, nodes attempt to avoid the reflection port and instead use the transmission port. In doing so, packets TRR is increased. In shufflenet TRR also reflects the output port utilisation. However, introducing the priority scheme decreases the output port utilisation dramatically thus resulting in an increased network congestion [6]. 5,1 6,1 7,1 5,2 6,2 7,2 Worst Case Scenario Best Case Scenario Figure 3. Shufflenet topology showing possible worst and best paths using prioritised routing 4. Results Here we present results for the TRR of shufflenet with respect to the system load. A random traffic model is used by generating packets based on predefined probability. The simulation is performed for 1000 iterations. Each test is carried out using a shufflenet with and without buffering (only one-level). In our simulation, we have made the following assumptions:- (i) Packets are processed synchronously by delaying them at the input ports so that they arrive at the router at the same time, (ii) packets are delayed while the router is processing the header containing the destination information, (iii) fixed packet size, and (iv) there are no packet loss as packets are deflected when there is no buffer available. Figure 4 shows the TRR versus system load for different shufflenet configurations employing deflection routing algorithm. For all cases, TRR drops as the system load

4 increases. This is because, there are less multiple paths when the system load increases. Moreover, as the system load increases the possibility of two packets arriving simultaneously at a node increases. Thus, both output ports of the router are used. As expected at low load (< ~0.17), larger configurations show higher TRR. Whereas at high load (> ~0.17) the TRR is higher for smaller shufflenet configurations. This is because, for larger shufflenet configurations there are more packets in the system, and therefore the possibility of finding non-congested paths decreases. T/R Ratio Shuffle Net Comparison - T/R Ratio vs System Load SN(2,3)-24N SN(2,4)-64N SN(2,5)-160N SN(2,6)-384N SN(2,7)-896N Figure 4. TRR vs. system load for different shufflenet configuration with deflection routing Figure 5 shows TRR versus system load when employing store-and-forward routing. As expected, the TRR profile is almost same as in Figure 4, since TRR does not dependent on buffer. Note that the deflection routing gives slightly better performance. This is because the average number of hops taken per packet is 25% higher than storeand-forward routing [7]. Indeed, TRR is directly proportional to the number of multiple paths that exists. Therefore, one can say that having a buffer has no effect on number of multiple paths. As in Figure 4, the decrease in the TRR as the load increases is due to an increased number of packets within the network, thus increasing the occurrence of deflection. Packets not being deflected also decrease the possibility of multiple hops. When a packet is deflected, it is at least k hops away from its destination, which makes it TYPE-M packet. T/R Ratio Shuffle Net Comparison (Buffer) - T/R Ratio vs SN(2,3)-24N SN(2,4)-64N SN(2,5)-160N SN(2,6)-384N SN(2,7)-896N Figure 5. TRR vs. system load - with store-and-forward routing 5. Conclusions In this paper we have introduced shufflenet topology employing TOAD based 2X2 OTDM Router. A simple priority scheme was developed where packets are routed via the transmission port instead of the reflection as much as possible. Simulation results showed that TRR > 1 is achievable over a wide range of system load able at system load for both deflection and storeand-forward routings. Higher TRR means that packet will observe lower crosstalk when propagating through the network. One drawback of prioritised scheme is that the output port (channel) utilisation is imbalanced as one of the ports has priority over the other. Other topologies (such as Gemnet [8]) are currently being considered in the same context. 6. Acknowledgement Firat Tekiner is financially sponsored by DTRI scholarship.

5 7. References 1- A. S. Acampora, A Multichannel Multihop Local Lightwave Network, Proc. IEEE GLOBECOM 87, pp , Nov A. S. Acampora, M. J. Karol, M. G. Hluchyj, Multihop Lightwave Networks: A New Approach to Achieve Terabit Capabilities, in Proc. ICC'88, vol. 1, pp , R. Gao, Z. Ghassemlooy, G. Swift and P. Ball, Simulation of All-Optical Time Division Multiplexed Router, SPIE Photonics West, USA, A. S. Acampora and Mark J. Karol, An Overview of Lightwave Packet Networks, IEEE Network, pp , Jan S. Banerjee, V. Jain, S. Shah, Regular Multihop Logical Topologies for Lightwave Networks, IEEE Communications Surveys, First Quarter M. J. Karol and S. Shaikh, A Simple Adaptive Routing Scheme for ShuffleNet Multihop Lightwave Networks, GLOBECOM 88, Conf. Rec., pp , Nov N. F. Maxemchuk, Comparison of Deflection and store-and-forward techniques in the MSN and Shuffle Exchange networks, Proc. IEEE INFOCOM 89, April J. Iness, S. Banerjee, and B. Mukherjee, GEMNet: A Genaralized, Shuffle Exchange-Based, Regular, Scalable, Modular, Multihop, WDM Lightwave Network, IEEE/ACM Trans. on Networking, vol. 3, no. 4, pp , Aug

Routing In the Logical Interconnection Topologies

Routing In the Logical Interconnection Topologies Firat Tekiner First Year PhD Student Zabih Ghassemlooy, Thompson Mark, Alkhayatt Samir Optical Communications Research Group School of Engineering, Sheffield