Transmission Rate Control Utilizing Chaotic Nature of Coupled Oscillators Yuri Takahashi†, Chisa Takano‡, Yusuke Sakumoto†, and Masaki Aida† † Graduate School of System Design, Tokyo Metropolitan University Hino-shi, Tokyo 191-0065, Japan Email: {takahashi-yuri, sakumoto, maida}@sd.tmu.ac.jp ‡ Graduate School of Information Sciences, Hiroshima City University Hiroshima 731-3194, Japan Email: takano@hcu.ac.jp Abstract—The TCP global synchronization problem is a phenomenon that packet losses or window control actions of different flows are synchronized. It causes degradation of the utilization of network bandwidth. RED is a well known approach to avoid this problem. However, discard of packets occurs in some probability even if they are in low-rate flow, and it might be unfair. In this paper, we propose new transmission rate control by using chaotic nature appearing in coupled oscillators. Our control can avoid the synchronization problem despite no use of random numbers and improve the network utilization. Keywords-coupled oscillators; chaos; TCP global synchro- nization I. I NTRODUCTION Nowadays, information networks support various activ- ities in the real world, and the networks are thought as a kind of important social infrastructure. Transmission Control Protocol (TCP) is used widely as a transport layer protocol, to realize reliably communications across a network. Highly-reliable communications are realized by TCP’s three functions: packet sequencing, re-transmission, and flow/congestion control by using window control. The received side sends an acknowledgment (ACK) packet if the forwarded packets are correctly received. Then the window size increases if the sending side receives the ACK packet. If the ACK packet is not received correctly, the sending side would recognize that congestion occurs in the networks. In this case, the sending side shrinks the window size and it decreases the amount of forward packets. Here, if different flows shrink their window size simultaneously, the amount of forwarded packets is extremely decreased, and it would cause degradation of throughput. This problem is called TCP global synchronization. Here, we explain the mechanism of TCP global synchro- nization by using TCP Reno as an example [1]. Let us consider multiple flows of which each of their round-trip time (RTT) is almost the same. If packet losses occur at the packet buffer in a router on the common path of their flow, the window sizes of these flows shrink simultaneously. Since RTT is the same, the procedure of window size control for the multiple flows would be synchronized. That is, the procedures of shrinking window size by half and increasing it gradually are synchronized over the multiple flows. As a result, window sizes of the multiple flows grow simultaneously, and they cause congestion again. This situation means inefficient use of the network. The state of the networks repeats very busy and almost vacant states. It causes degradation of the utilization of network bandwidth. If flows are asynchronous and behaviors of window sizes are mutually independent, we can process more amount of data in the network. Random Early Detection (RED) is known as a way to solve TCP global synchronization. RED can avoid this problem by dropping packets randomly before router buffer will be full of packets [2]. So it can prevent simultaneous packet losses belonging to many flows. Since the cause of congestion is flows with many packet arrivals, we want to shrink the window size of such a higher-rate flow. The strategy of RED is to provide congestion avoidance by dropping arrival packet randomly when the buffer utilization is high. This is because higher-rate flow generates many packets. That is, the probability that the randomly dropped packet is belonging to higher-rate flow is relatively high. Concretely, RED is explained as follows; Action to the arriving packets depends on the average queue size of buffer. If the average is less than the predefined threshold, the arriving packet does not dropped. If the average is greater than the threshold, the arriving packet is dropped with a certain probability. The probability is proportional to the difference between the average queue length and the threshold. In addition to that, RED has another bigger threshold. If the average queue size exceeds it, arriving packets are dropped with probability 1. In this way, RED is trying to implement efficient congestion avoidance by dropping packets randomly. This is because the probability that the randomly dropped packets are belonging to the flows that cause the congestion is relatively high . However, it might drop low-rate flow packets in some cases, because the dropped packets are chosen at random. In particular, it is the case when there are a lot of low-rate flows. Let us consider 100 low-rate flows and only one high-rate flow, and assume that the high-rate flow has 100 times rate of the 2012 Fourth International Conference on Intelligent Networking and Collaborative Systems 978-0-7695-4808-1/12 $26.00 © 2012 IEEE DOI 10.1109/iNCoS.2012.69 126