Kousuke Nogami

Robust available bandwidth estimation against dynamic behavior of packet scheduler in operational LTE networks

We shed new light on the mechanism behind how the dynamic behavior of a packet scheduler at the link layer in mobile networks degrades the accuracy of conventional available bandwidth (i.e., unused capacity of an end-to-end path) estimation methods that use a probing packet train (i.e., a set of multiple probing packets). Most of the conventional methods, which were originally designed for wired networks, estimate available bandwidth at the receiver by detecting changes of the observed queuing delays of probing packets. They utilize a microscopic approach in which they check the difference of the queuing delay of each packet on a packet-by-packet basis in order to detect the queuing delay changes. We found that the dynamic behavior of a packet scheduler at the link layer dramatically disturbs the queuing delays observed at the receiver. The disturbed queuing delays make it tremendously difficult for the conventional microscopic approach to detect changes of the delays, resulting in degraded estimation accuracy.

A method for reducing TCP latency due to consecutive packet losses in mobile networks

In this paper, we propose a method for reducing the transmission control protocol (TCP) latency due to consecutive packet losses in mobile networks. Mobile networks are becoming increasingly popular and used in the same sense as fixed networks. However, insufficient attention is directed toward the problem with mobile networks, i.e., long latency due to an inappropriate TCP retransmission mechanism. In the proposed method, a sender detects the start and end of burst packet losses and retransmits right after the endpoint. To detect the endpoint, the sender periodically sends a small probe packet during the waiting state. By the proposed method, latency decreased without increasing traffic load in 70% of sample cases compared with the conventional method.

Experimental comparison of machine learning-based available bandwidth estimation methods over operational LTE networks

We propose PathML, an available bandwidth (i.e., unused capacity of an end-to-end path) estimation method based on a data-driven paradigm that uses machine learning with a large amount of data. An experiment over an operational LTE network was performed to compare our method with prior work.

A method for correcting minimum RTT of TCP congestion control in mobile networks

In mobile networks such as LTE and 3G, there is a problem in that the transmission control protocol (TCP) cannot sufficiently utilize network bandwidth and TCP throughput decreases. One cause of this degraded TCP throughput is the underestimation of minimum round trip time (RTT) in TCP congestion control algorithms. Handovers between mobile networks might increase propagation delay, so it is important to increase the minimum RTT appropriately. However, conventional methods cannot increase the minimum RTT because the minimum RTT is updated when a shorter RTT is observed. This causes many delay-based TCP variants to decrease the congestion window size, which in turn decreases TCP throughput. In this paper, we propose a method that revises the minimum RTT in mobile networks based on information available at the TCP sender node. Simulation results indicate that the proposed method can appropriately revise the minimum RTT.

Analysis of Network Throughput Dynamics with Viscoelastic Material Model

Network throughput of the best-effort networks as typified by the Internet and wireless networks is often insufficient for multimedia communication, and it spoils the quality of communication. Therefore, it is important to analyze a dynamic characteristics of the throughput when multimedia traffic is injected. We point out that the dynamics of the throughput has the viscosity and the elasticity because of an interaction among traffic passing through the network. We also introduce a viscoelastic material model to analyze the throughput. A total of 120 simulations were performed with an ns-2 simulation model. The results of simulations showed that the dynamics of the throughput does indeed have the viscoelasticity. Furthermore, we found that our proposed model can be used to analyze the throughput dynamics with a fit rate of about 90% on a wired network, and about 80% on a wireless one. This fact proved that there was an analogous relationship between variation of throughput and deformation of the viscoelastic material. We believe that we have developed a novel science field of Network Rheology.

Node inquiry function for strategic front-end placement

In a front-end system, mirror servers called front-ends are deployed on nodes within the Internet to enhance the quality of on-line services. An on-line service provider can control its service quality by changing the number of front-ends. In addition to the number, the placement of front-ends is also important for the provider to maximize the profitability. This paper proposes a node inquiry function (NIF) that provides on-line service providers with strategic front-end placement plans. The NIF-tailored placement plan enables a service provider to obtain the best possible front-end distribution that maximizes investment efficiency. The NIF also helps them to find those heterogeneous nodes that can run their front-ends. The placement algorithm calculates in polynomial time and an experiment shows better approximation than a greedy approach.