Mikyung Kang

Heterogeneous Cloud Computing

Current cloud computing infrastructure typically assumes a homogeneous collection of commodity hardware, with details about hardware variation intentionally hidden from users. In this paper, we present our approach for extending the traditional notions of cloud computing to provide a cloud-based access model to clusters that contain a heterogeneous architectures and accelerators. We describe our ongoing work extending the Open Stack cloud computing stack to support heterogeneous architectures and accelerators, and our experiences running Open Stack on our local heterogeneous cluster test bed.

Energy consumption scheduler for demand response systems in the smart grid

This paper presents a design and evaluates the performance of a power consumption scheduler in smart grid homes or buildings, aiming at reducing the peak load in them as well as in the system-wide power transmission network. Following the task model consist of actuation time, operation length, deadline, and a consumption profile, the scheduler linearly copies the profile entry or maps a combinatory vector to the allocation table one by one according to the task type, which can be either preemptive or nonpreemptive. The proposed scheme expands the search space recursively to traverse all the feasible allocations for a task set. A pilot implementation of this scheduling method reduces the peak load by up to 23.1% for the given task set. The execution time, basically approximated by (The equation is abbreviated), where M, N(subscript NP), and N(subscript P) are the number of time slots, nonpreemptive tasks, and preemptive tasks, respectively, is reduced almost to 2% taking advantage of an efficient constraint processing mechanism which prunes a search branch when the partial peak value already exceeds the current best. In addition, local peak reduction brings global peak reduction by up to 16% for the home-scale scheduling units without any global coordination, avoiding uncontrollable peak resonance.

GPU Passthrough Performance: A Comparison of KVM, Xen, VMWare ESXi, and LXC for CUDA and OpenCL Applications

As more scientific workloads are moved into the cloud, the need for high performance accelerators increases. Accelerators such as GPUs offer improvements in both performance and power efficiency over traditional multi-core processors, however, their use in the cloud has been limited. Today, several common hypervisors support GPU passthrough, but their performance has not been systematically characterized. In this paper we show that low overhead GPU passthrough is achievable across 4 major hypervisors and two processor microarchitectures. We compare the performance of two generations of NVIDIA GPUs within the Xen, VMWare ESXi, and KVM hypervisors, and we also compare the performance to that of Linux Containers (LXC). We show that GPU passthrough to KVM achieves 98 -- 100\% of the base systems performance across two architectures, while Xen and VMWare achieve 96 -- 99\% of the base systems performance, respectively. In addition, we describe several valuable lessons learned through our analysis and share the advantages and disadvantages of each hypervisor/GPU passthrough solution.

An Efficient Scheduling Scheme on Charging Stations for Smart Transportation

This paper proposes a reservation-based scheduling scheme for the charging station to decide the service order of multiple requests, aiming at improving the satisfiability of electric vehicles. The proposed scheme makes it possible for a customer to reduce the charge cost and waiting time, while a station can extend the number of clients it can serve. A linear rank function is defined based on estimated arrival time, waiting time bound, and the amount of needed power, reducing the scheduling complexity. Receiving the requests from the clients, the power station decides the charge order by the rank function and then replies to the requesters with the waiting time and cost it can guarantee. Each requester can decide whether to charge at that station or try another station. This scheduler can evolve to integrate a new pricing policy and services, enriching the electric vehicle transport system.

An energy-efficient real-time scheduling scheme on dual-channel networks

The recent evolution of wireless sensor networks have yielded a demand to improve energy-efficient scheduling algorithms and energy-efficient medium access protocols. This paper proposes an energy-efficient real-time scheduling scheme that reduces power consumption and network errors on dual channel networks. The proposed scheme is based on a dynamic modulation scaling scheme which can scale the number of bits per symbol and a switching scheme which can swap the polling schedule between channels. Built on top of EDF scheduling policy, the proposed scheme enhances the power performance without violating the constraints of real-time streams. The simulation results show that the proposed scheme enhances fault-tolerance and reduces power consumption.

Design and Development of a Run-Time Monitor for Multi-Core Architectures in Cloud Computing

Cloud computing is a new information technology trend that moves computing and data away from desktops and portable PCs into large data centers. The basic principle of cloud computing is to deliver applications as services over the Internet as well as infrastructure. A cloud is a type of parallel and distributed system consisting of a collection of inter-connected and virtualized computers that are dynamically provisioned and presented as one or more unified computing resources. The large-scale distributed applications on a cloud require adaptive service-based software, which has the capability of monitoring system status changes, analyzing the monitored information, and adapting its service configuration while considering tradeoffs among multiple QoS features simultaneously. In this paper, we design and develop a Run-Time Monitor (RTM) which is a system software to monitor the application behavior at run-time, analyze the collected information, and optimize cloud computing resources for multi-core architectures. RTM monitors application software through library instrumentation as well as underlying hardware through a performance counter optimizing its computing configuration based on the analyzed data.

An efficient bandwidth management scheme for a hard real-time fuzzy control system based on the wireless LAN

This paper proposes and analyzes bandwidth allocation and reclaiming schemes on wireless media to enhance the timeliness of the real-time messages and accordingly the correctness of fuzzy control decision. Bandwidth allocation scheme generates efficient round robin polling schedule represented as a capacity vector by directly considering the deferred beacon problem. The resource reclaiming scheme reassigns unused slot time to non-real-time traffic by extending the collision period without violating the hard real-time guarantee. The simulation results show that the proposed scheme can not only enhance the schedulability of wireless network by up to 18% but also give more bandwidth to the non-real-time traffic up to 5.3%, while the resource reclaiming scheme can maximally improve the achievable throughput by 11% for the given stream set.

Programming Models and Development Software for a Space-Based Many-Core Processor

The Maestro processor is a 49-core many-core processor for space based on the TILE64 architecture and implemented in rad-hard-by-design technology by Boeing. In this paper we discuss the programming models for Maestro, the implications of the programming model on fault tolerance and flight software, and the software development tools that have been developed for Maestro. The software described here is experimental development software that allows application and algorithm evaluation on the architecture, but we believe this software can be used as the basis for flight software. The software includes libraries, performance analysis and optimization tools, and compilers. While this work was done on the Maestro chip, the principles discussed can be applied to any multi-core or many-core processor.

Data collection scheme for two-tier vehicular sensor networks

This paper proposes a data collection scheme for the twotier vehicle sensor network, which is comprised of high-speed WLANs surrounded by the ubiquitous cellular network, aiming at improving the accuracy and speed of event detection out of bunch of sensor data. Each vehicle reports periodically summary data including its location, timestamp, and alarm data via the cellular network, while it sends whole set, entering the WLAN range. For event detection, the information server locates the vehicles whose data must be further investigated according to the legacy spatio-temporal query or the trajectory comparison method. The WLAN can give precedence to the vehicle which might have the important data by operating two different frequencies. We are currently refining this framework and planning to test in Jeju area.

Design for Run-Time Monitor on Cloud Computing

Cloud computing is a new information technology trend that moves computing and data away from desktops and portable PCs into large data centers. The basic principle of cloud computing is to deliver applications as services over the Internet as well as infrastructure. A cloud is the type of a parallel and distributed system consisting of a collection of inter-connected and virtualized computers that are dynamically provisioned and presented as one or more unified computing resources. The large-scale distributed applications on a cloud require adaptive service-based software, which has the capability of monitoring the system status change, analyzing the monitored information, and adapting its service configuration while considering tradeoffs among multiple QoS features simultaneously. In this paper, we design Run-Time Monitor (RTM) which is a system software to monitor the application behavior at run-time, analyze the collected information, and optimize resources on cloud computing. RTM monitors application software through library instrumentation as well as underlying hardware through performance counter optimizing its computing configuration based on the analyzed data.