Ken Igarashi

Dynamic resource allocation and power management in virtualized data centers

We investigate optimal resource allocation and power management in virtualized data centers with time-varying workloads and heterogeneous applications. Prior work in this area uses prediction based approaches for resource provisioning. In this work, we take an alternate approach that makes use of the queueing information available in the system to make online control decisions. Specifically, we use the recently developed technique of Lyapunov Optimization to design an online admission control, routing, and resource allocation algorithm for a virtualized data center. This algorithm maximizes a joint utility of the average application throughput and energy costs of the data center. Our approach is adaptive to unpredictable changes in the workload and does not require estimation and prediction of its statistics.

Care-of Prefix Routing for Moving Networks

In future, mobile terminals may be linked in various types of local network where the whole network is moving. Mobile networks will need to provide global connectivity to such moving networks and manage their mobility. A moving network consists of mobile terminals and a mobile router, which acts as the gateway to the mobile network. To manage the mobility of the moving network, it is important to minimize the packet overhead, to optimize routing, and to reduce the volume of handoff signals over the mobile network and air interface. This paper proposes a new routing mechanism using hierarchical mobile network prefix assignment, home agent concatenation, hierarchical address management, and hierarchical re-routing. In hierarchical mobile network prefix assignment, a mobile router is assigned a mobile network prefix, which is used as a prefix when allocating the location addresses of mobile terminals in the moving network, so allowing them to be managed in a hierarchical manner. Home agent concatenation limits the number of home agents which need to be updated during handoff by enabling one home agent hold information relating to others, while hierarchical address management minimizes the volume of handoff signals by managing the location addresses of all mobile terminals in a hierarchical manner. Hierarchical re-routing introduces a local anchor router in order to localize handoff and to optimize routing. Simulation results show that our proposed routing method is better than the conventional solutions in terms of efficiency of data transmission including data transmission delay, and handoff performance.

In-network live snapshot service for recovering virtual infrastructures

Infrastructure as a Service (IaaS) has become an increasingly popular type of service for both private and public clouds. The virtual infrastructures that enable IaaS support multitenancy by multiplexing the computational resources of data centers and result in substantial reductions in operational costs. Since hardware and software failures occur on a routine basis in large-scale systems, it is imperative for cloud providers to offer various failure recovery options for distributed services hosted on such infrastructures. In this article we present GENI-VIOLIN, a new cloud capability that can checkpoint a stateful distributed service while incurring very low overhead. The unique aspect of GENI-VIOLIN compared to previous work is that GENI-VIOLIN exploits programmable OpenFlow switches to provide checkpointing services in the network, thereby requiring minimal changes to the end host virtualization framework. We have developed a prototype of GENI-VIOLIN using the GENI infrastructure, and have demonstrated GENI-VIOLINs checkpoint and restore capability across multiple GENI sites.

Flight size auto tuning for broadband wireless networks

The emergency of broadband wireless access technologies enables seamless Internet access in mobile environments. To fully utilize the broadband wireless bandwidth, it is critical to design a congestion control mechanism that can adjust the congestion window size according to the dynamics of the wireless network condition. A congestion window size that does not react properly to the degradation of channel quality injects excessive data segments into the network. This increases round trip time (RTT) and ultimately may cause packet losses. In the reverse setting, when the channel quality improves drastically, the congestion window should react fast enough to take advantage of the available bandwidth. This paper proposes Flight Size Auto Tuning (FS-AT) as a new TCP variant. FS-AT adjusts the transmission rate according to the estimated available throughput and fluctuations in throughput to mitigate the network congestion while maintaining high network utilization. Moreover, FS-AT provides mechanisms to differentiate between the losses caused by channel errors and those by network congestion. Accordingly, the proposed method avoids unnecessary throughput reduction in case of channel errors, a well-known impediment of standard TCP schemes. FS-AT estimates the current available bandwidth from the received acknowledgments without requiring any modification at the receiver. FS-AT is implemented on Linux and the effectiveness of FS-AT is validated by the measured performance of FS-AT over HSDPA and the simulated performance over a fixed network.

Care-of-prefix routing for moving networks in mobile IP network

The future ubiquitous network will serve so many mobile terminals that it is extremely important to control them efficiently. One useful approach is to group terminals having similar movement characteristics and manage them in units of groups. Another important issue is the mobility management of moving networks, such as a network on a train or in a car, or a personal area network. Moving networks may be defined for a variety of situations and can lead to a lot of attractive applications. Moving network mobility support is indeed one of the most interesting research topics. In this paper, we clarify the difference between host mobility support and the conventional moving network mobility support, propose a mechanism for moving network mobility support and shows it is better than the conventional ones.

Mobility aware TCP congestion control

We examine the performance of TCP over wireless networks. In wireless environments, packets may be lost until a mobile terminal completes registration with an access router. The technique of post registration handover is expected to reduce this loss since the mobile terminal sends a binding update to the previous care-of address after handover and packets are forwarded to the mobile terminal. However, if several packets have been lost, TCP throughput may become worse than not forwarding packets. Upon receiving the forwarded packets, the mobile terminal returns duplicate ACKs, and these ACKs force the sender into the congestion control phase. In New-Reno TCP, the senders transmission rate is extremely limited during the period. As a result, the throughput is reduced significantly. We solve the problem by using either TCP flow control or congestion control functions according to the number of packets lost. Results from simulations show that our proposal keeps the throughput high even if many packets are lost. Our study makes clear the need for a reliable transport protocol that is handover-aware and suggests how to adapt the protocol to yield better performance in wireless environments.