Tomoyoshi Sugawara

Heuristics and Evaluations of Energy-Aware Task Mapping on Heterogeneous Multiprocessors

On various heterogeneous multiprocessor platforms, it is necessary to optimize the throughput and the energy consumption. The problem of optimally mapping tasks onto a set of given heterogeneous processors for minimum overall completion time has been known, in general, to be NP-complete. The energy consumption of a task may be very different in heterogeneous processors. However, if the most energy-efficient processor for each task is always chosen in the task mapping, the completion time of a set of tasks may grow wildly in the worst case. DVS (Dynamic Voltage Scaling) technique is currently available in a larger number of processors to effectively reduce dynamic power dissipation and consequently to save a proportion of total energy consumption, but meanwhile the execution time of a task running in lower voltage definitely becomes longer. Hence, the task mapping problem in terms of time, energy and voltage turns more complicated and harder to solve along with the heterogeneity. Moreover, today most processors only support discrete DVS and thus the optimization problem tends to be Integer Linear Programming problems for which, as we know, there is no polynomial time algorithm unless P = NP. In this paper we formulate and study the optimization problem of reducing overall completion time and the total energy consumption, and then some heuristics, which are experimentally evaluated and compared.

TCP-Migration with Application-Layer Dispatching: A New HTTP Request Distribution Architecture in Locally Distributed Web Server Systems

A cluster-based server system is a developing technology that could achieve high scalability by using several dispatchers, such as layer-4 or layer-7 switches, to appropriately distribute requests from clients. Many recent Web server systems have been developed as cluster systems, but such systems are so complicated that important information for appropriate distribution decisions is in higher layer (i.e., application layer, or layer-7). Although the L7 switches are appliances that can redirect requests by examining the application-layer information, it is difficult to update or modify their distribution algorithms. This paper proposes a novel architecture based on TCP-migration mechanism that provides complete redirection (displacement) of a TCP session from a dispatcher to Web servers. The key idea is physical separation of L7 switch functionality: packet forwarding and request dispatching mechanisms. With NAT mechanism on the L3 switch and sophisticated management of virtual private IP addresses on the cluster servers, the dispatcher is released from relaying or translating both in-bound and out-bound TCP packets after the request has been redirected. This architecture can achieve greater flexibility because the forwarding is performed fast by hardware (i.e., the switch), and the dispatching is managed by software (i.e., application servers). We have designed and implemented this mechanism on Linux 2.4 kernel and evaluated its performance. The experimental results show that the overhead for handling multiple virtual IP addresses is almost negligible. Furthermore, the overhead with TCP-migration by using mini_httpd server and wget client is approximately 1 ms, regardless of the reply size, on 3.06 GHz Xeon machines

GMPLS with Interlayer Control for Session-Uninterrupted Disaster Recovery across Distributed Data Centers

We propose a session-uninterrupted disaster recovery system using a novel session migration technique as a GMPLS application. Existing disaster recovery systems have a problem of a service interruption. The session migration based on an interlayer control of GMPLS, VLAN change-over, and process migration, maintains continuous TCP service between a user and a virtualized server, even when the service migrates from a primary data center to a backup one. We developed a prototype system and showed that BoD (bandwidth on demand) by GMPLS improved the recovery time from 80.10 sec to 9.85 sec, during transmitting a process data of 40MByte.

An Optimal Power Management for Stationary Multiprocessor Systems

Power consumption has become a critical issue in designing computer systems. Dynamic power management is an approach that aims to reduce power consumption at system level by selectively placing components into low power states. In this paper, a power management for multiprocessor systems is proposed to optimally reduce the power consumption of processors. The key feature of the proposed power management is that sleep tasks, virtual tasks that merely switch processors into low power states, are generated and injected into the original task traffic in advance. High priorities are given to real tasks and sleep tasks are serviced only on necessity. In order to avoid unnecessary state transfers, no preemption is allowed. The goal is to appropriately switch idle processors into low power states and meanwhile minimize the average response time of real tasks. A probabilistic policy towards the goal is also introduced. The analysis of the probabilistic policy is based on such an queueing model: there are multiple processors and a global queue, the system is in equilibrium, i.e. a stationary traffic and a stable queue, except two priorities tasks are same and non-preemptive. Based on the queueing model, the optimality of the proposed power management is discussed and analyzed.

μ‐PULSER: A reactive operating system for personal robots

When the “personal robot,” which is one of the future goals of the autonomous mobile robot, is viewed from the standpoint of computer control, there must be a reactive mechanism that reflects the sudden change of the sensor output on the emergency processing task. When the personal robot is viewed as a robot owned by the individual and to be used for the individual purposes, it must be highly expandable and flexible so that the functions and the hardware configuration can be varied according to the purpose. To support such a robot, the authors developed a multithread operating system μ-PULSER. μ-PULSER provides a high-speed synchronization mechanism among threads, which helps to realize the reactive control of the robot. The high-speed communication mechanism is provided among the threads, by utilizing the forementioned synchronization mechanism. By minimizing the kernel, it is made possible to reconfigure dynamically the system by exchanging and adding the servers. This paper discusses the design and implementation of μ-PULSER. It is shown by actual measurement and evaluation that the synchronization and the communication mechanism among the threads have sufficiently high speeds to meet the requirement of robot control.

Virtualized server infrastructure for resilient voice communication service

The huge earthquake struck Japan on 11th March, 2011, caused a massive congestion of call attempts from mobile phones that resulted in only 5% of accepted connections due to the congestion control by telephone companies. It is an emergent issue to improve resiliency of communication service in anticipation of future disasters and thus communication service infrastructure necessitates the flexibility of its capacity. In this paper, we introduce server virtualization technology to provide a flexible communication service infrastructure and design a communication service controller that provisions additional capacity by virtual machines in response to increased call attempts from mobile phones after a disaster. The communication service controller is designed with constraint programming and performance/availability estimations for deciding the optimum virtual machine placement according to network operators instructions. Through an experimental disaster test, we confirm the capacity of communication service is increased five-fold by virtual machine provisioning with half an hour latency.

Architecture of a parallel computer Cenju-4

This paper describes the architecture and the evaluation results of a parallel computer Cenju-4. Cenju-4 supports two memory architectures: distributed memory with user-level message passing communication and distributed shared memory with cache-coherent nonuniform memory access (cc-NUMA) feature. Cenju-4 system consists of from 8 to 1024 nodes connected by a multistage network which has multicast, synchronization, and gather functions. Cenju-4 adopts a Mach micro kernel based operating system, which provides several services for parallel processing. We attained 5.5 psec communication latency and 168 Mbytes/sec communication throughput an message passing communication.