Akiyoshi Sugiki

Power-Saving in Large-Scale Storage Systems with Data Migration

We present a power-saving method for large-scale distributed storage systems. The key idea is to use virtual nodes and migrate them dynamically so as to skew the workload towards a small number of disks while not overloading them. Our proposed method consists of two kinds of algorithms, one for gathering or spreading virtual nodes according to the daily variation of workloads so that the active disks are reduced to a minimum, the other for coping with the changes in the popularity of data over a longer period. For this dynamic migration, data stored in virtual nodes are managed by a distributed hash table. Furthermore, to improve the reliability as well as to reduce the migration cost, we also propose an extension of our method by introducing a replication mechanism. The performance of our method is measured both by simulation and a prototype implementation. From the experiments, we observed that our method skews the workload so that the average load for the active physical nodes as a function of the overall capacity is 67%. At the same time, we maintain a preferred response time by setting a suitable maximum workload for each physical node.

An Extensible Cloud Platform Inspired by Operating Systems

Virtualization has changed the ways computation is done, especially in utility and cloud computing. While several virtualization cloud platforms have emerged to provide interfaces between users and data centers, traditional operating systems (OSes) have defined interfaces between applications and hardware. We argue that virtualization cloud platforms and OSes can share many problems, thus, many techniques developed for OSes can be applied to such platforms. We have implemented a prototype of an OS-like middleware that adopted a micro kernel design, had resource abstraction achieved by objects and functions, and offered a scripting environment as a programming interface. The experimental results demonstrated that it had the ability to achieve elasticity and high-availability VMs. We also carried out qualitative comparisons with other middleware to clarify our research position.

An integrated management framework for virtual machines, switches, and their SDNs

The paradigm of software-defined networks (SDNs) and one of its enabling technologies, OpenFlow, have been widely welcomed by researchers and practitioners because of their flexibility in networking. Most cloud operators are attracted by the ability to “virtualize” networks at data centers, which is one of the most appealing out of various promising applications. How virtual networking is implemented in SDNs still remains to be solved in future research despite such enthusiasm to achieve virtualization. When researchers are involved in SDN investigations, they soon notice the tediousness of setting up the study environment, which consists of virtual machines (VMs), switches, and SDN controllers. Therefore, they need to spend a great deal of effort and time on non-essential parts of their studies. We present Kumoi in this paper, which is an integrated development framework for VMs, switches, and SDNs. Kumoi enables us to rapidly deploy VMs along with OpenFlow networking. Kumoi is not just an OpenFlow controller, but it also provides the ability to manage VMs in a single environment. We believe Kumoi could accelerate research on VMs and SDNs.

A P2P Approach to Scalable Network-Booting

Network-booting is widely adopted in universities that have to maintain many client computers. In conventional network-booting systems, the primary bottleneck is the disk image distribution servers and the network to these servers. To eliminate this bottleneck, peer-to-peer (P2P) methods must work. However, existing P2P methods, including BitTorrent, do not work well for network-booting because they are highly optimized for distributing an entire large file, while network-booting requires certain parts of a large file. In the present study, aiming to solve the problems described above, we describe a new network-booting system that uses a P2P method. In our P2P-based network-booting system, a client node receives disk blocks not only from central distribution servers but also from the other client nodes that already have the demanded disk blocks. To the best of our knowledge, our network-booting system is the first effective implementation of a network-booting system that uses a P2P method in a local area network. Unlike conventional P2P systems, the proposed network-booting system can deal with demanded parts of a large disk image. We performed experiments with 112 client nodes in real classrooms on a university campus. The results of the experiments show that our implementation scales well as the number of client nodes grows.

Self-Stabilizing Passive Replication for Internet Service Platforms

We present a self-stabilizing passive replication mechanism for Internet service platforms. The proposed system based on this mechanism aims to provide services despite simultaneous failure of the majority of nodes. To achieve this objective, we use the self- stabilizing consensus algorithm introduced by Dolev et al., instead of a majority-based consensus algorithm such as Paxos. Although our system allows temporary illegal behavior, it eventually converges on the desired state without interruption of services. Thus, the proposed system is useful for providing services that require high availability, but not strict consistency, a prime example of which is an Internet bulletin board for communication during large-scale natural disasters. We implemented a prototype and evaluated it with experiments to demonstrate availability through various patterns of failures.

Dynamic grid quorum: A novel approach for minimizing power consumption without data migration in grid quorums

In this paper, we present the dynamic grid quorum, a technique for reducing the power consumption of large-scale storage systems. The key idea is to skew the workload towards a small number of quorums by means of a novel optimization algorithm. Moreover, this system allows reconfiguration by exchanging nodes without any data migration so as to reallocate high-capacity nodes to busier quorums. We demonstrate that the dynamic grid quorum saves, on average, 6–12% energy when compared with the static configurations.

A platform for cooperative server backups based on virtual machines

We present a virtual machine-based peer-to-peer platform that allows many Internet services to back up their services cooperatively. The goal of our platform is to provide a highly-available and service-independent solution that is cost-effective for smaller, independent service providers. Use of virtual machines makes it possible to encapsulate the complete service state and to share physical hosts. A multicast protocol guarantees that service state is replicated reliably on multiple physical hosts so that in case of failures, recent state can be recovered. We implemented a prototype and evaluated it by experiments to show that our design can adapt to dynamic host changes and evaluate the runtime and failure recovery performance possible with Xen and SBUML virtual machines.

Dynamic grid quorum: a reconfigurable grid quorum and its power optimization algorithm

Power-saving has become a central issue for well-configured SOC platforms. In particular, as a high percentage of the total energy is used by the storage systems, the cost effectiveness of data management is equally as important as reliability and availability. To address this issue, we propose the dynamic grid quorum as a method for reducing the power consumption of large-scale distributed storage systems. The basic principle of our approach is to skew the workload toward a small number of quorums. This can be realized using the following three techniques. First, our system allows reconfiguration by exchanging nodes without any data migration, so that high-capacity nodes can be reallocated to busier quorums. Second, for more effective skewing of the workload, we introduce the notion of dual allocation, which makes it possible to consider two distinct allocations in the same grid for write and read quorums. Finally, we present an optimization algorithm to find a pair of a strategy and an allocation of nodes, which minimizes power for a given system setting and its workload. We also demonstrate that the dynamic grid quorum saves, on average, 14–25% energy compared with static configurations, when the intensity of the total workload changes.