Modern deep learning frameworks support a variety of hardware, including CPU, GPU, and other accelerators, to perform computation. In this paper, we study how schedule jobs over such interchangeable resources - each with different rate computation optimize performance while providing fairness among users in shared cluster. We demonstrate theoretically empirically that existing solutions their straightforward modifications poorly the presence resources, which motivates design implementation...

Wireless sensor networks are becoming more widely used in various fields. Localization is a crucial and essential issue for network applications. In this paper, we propose simple, distributed low cost localization algorithm which approaches the benefits of range-free schemes. The proposed utilizes location information neighbor anchor nodes to estimate intersection points anchors' communication ranges. Then, by using RSS (Received Signal Strength)-comparison, it compares received signals from...

Reducing costs plays a crucial role in building and operating data centers. Internet service providers such as Facebook Google spend billions of dollars on capacity expansion operations their global Traditionally, planning for centers is done separately from operational management, which incurs inefficiency. In fact, management has significant impacts planning. Motivated by this gap, we propose framework that jointly optimizes both sustainable participating demand response programs....

Electric vehicle (EV) charging and discharging management is an essential component of the smart grids, especially in microgrids due to high dependence on local energy sources. In order provide EV users with freedom charge travel, we propose a dynamic scheduling algorithm that does not require inform their time control center. The proposed serves requests from EVs best-effort manner while minimizing cost guaranteeing grid related constraint microgrid. Specifically, formulate stochastic...

We propose an improved TDoA (Time Difference of Arrival) localization scheme based on PSO (Particle Swarm Optimization) in UWB (Ultra Wide Band) systems. The proposed is composed two steps: the re-estimation parameters and re-localization tag position. In both steps, algorithm employed to improve performance. first step, re-estimates obtained by traditional reduce estimation error. second with estimated re-localizes minimize location Simulation results show that achieves better performance...

The increasing number of electric vehicles (EVs) will have a significant impact on the demand response (DR) in smart grids (SGs). How to encourage EV users voluntarily join vehicle-to-grid (V2G) market has been one challenges SGs engineers and scientists. On other hand, dealing with randomness renewable distributed energy resources (DERs) output is critical for microgrids maintain their efficiency reliability. In this work, charging discharging scheduling scheme (DCD) EVs proposed provide...

Since the first arrival path may not be strongest of UWB (Ultra Wide Band) multipath channels, this makes ToA (Time-of-Arrival) estimation becomes a challengeable issue. Furthermore, because ultra bandwidth received signal, CS (Compressed Sensing) theory is employed to reduce complexity caused by very high Nyquist sampling rate in receivers. In paper, we propose scheme provides precise performance, while exploiting benefits CS-based Simulation results show that proposed can outperform other...

Motivated by the proliferation of heterogeneous processors such as multi-core CPUs, GPUs, TPUs, and other accelerators for machine learning, we formulate a novel multiinterchangeable resource allocation (MIRA) problem where some resources are interchangeable. The challenge is how to allocate interchangeable users in sharing system while maintaining desirable properties incentive, Pareto efficiency, envy-freeness. In this paper, first show that existing algorithms, including Dominant Resource...

Multi-timescale electricity markets augment the traditional market by enabling consumers to procure in a futures market. Heavy power consumers, such as cloud providers and data center operators, can significantly benefit from multi-timescale purchasing some of needed ahead time at cheaper rates. However, energy procurement strategy for centers becomes challenging problem when real world dynamics, spatial diversity uncertainties renewable energy, IT workload, price, are taken into account. In...

Heavy power consumers, such as cloud providers and data center operators, can significantly benefit from multi-timescale electricity markets by purchasing some of the needed ahead time at cheaper rates. However, energy procurement strategy for centers in becomes a challenging problem when real world dynamics, spatial diversity uncertainty renewable energy, IT workload, price, are taken into account. In this paper, we develop algorithms geo-distributed that utilize to minimize cost. We...

GPUs are considered as the accelerators for CPUs. We call these applications GPU applications. Some machine learning frameworks like Tensorflow support their (ML) jobs running either on CPUs or GPUs. Nvidia claims that Titan K80 12GB can speed up 5-10x average. Although offer advantages performance, they very expensive. For example, a roughly costs $4000 while an Intel Xeon E5 Quad cores $350. The coexist of traditional CPU and urges cloud computing operators to build hybrid CPU/GPU...

Even though batch, interactive, and streaming applications all care about performance, their notions of performance are different. For instance, while the average completion time can suffciently capture a throughout-sensitive batch-job queue (TQ) [5], interactive sessions form latencysensitive queues (LQ): each LQ is sequence small jobs following an ON-OFF pattern. these [7], individual times or latencies far more important than throughput LQ. Indeed, existing "fair" schedulers inherently...

Primary Frequency Control (PFC) is a fast acting mechanism used to ensure high-quality power for the grid that becoming an increasingly attractive option load participation. Because of speed requirement, PFC requires distributed control laws be instead more centralized design. Current designs assume costs at each geographic location are independent. Unfortunately many networked systems such as cloud computing, decisions made among locations interdependent and therefore require coordination....

UWB(Ultra Wide Band) 다중경로 채널에서 첫번째 경로를 통해 수신되는 신호가 가장 큰 아닐 경우가 종종 있으며, 이러한 경우 ToA(Time-of-Arrival) 추정의 정밀도를 유지하는 것은 매우 어려운 문제가 된다. 또한 UWB 신호의 초광대역 특성상 동기식 시스템을 구현할 수신기는 높은 표본화율을 이용해 신호를 수신해야 하기 때문에 복잡도가 증가되는데, 압축센싱(Compressed Sensing) 이론을 이용함으로써 시스템의 복잡도를 효율적으로 낮출 수 있다. 이에 본 논문은 압축센싱 기반의 수신기의 장점을 이용하면서도 정밀 추정성능을 제공할 있는 개선된 ToA 추정 기법을 제안한다. 모의실험 결과를 광범위한 신호대잡음비 환경에서 제안된 기법이 다른 저복잡도 기법들의 성능보다 우수함을 확인하였다. Since the first arrival path may not be strongest of multipath channels, this makes...

Primary Frequency Control (PFC) is a fast acting mechanism used to ensure high-quality power for the grid that becoming an increasingly attractive option load participation. Due speed requirement and other considerations, it often desirable have distributed control laws. Current PFC designs assume costs at each geographic location are independent. However, many networked systems, such as those cloud computing, interdependent across locations therefore need coordination. In this paper, laws...

Primary frequency control (PFC) is a fast- acting mechanism used to ensure high-quality power for the grid that becoming an increasingly attractive option load participation. Due speed requirements and other considerations, it often desirable have distributed laws. Current PFC designs assume costs at each geographic location are independent. However, many networked systems, such as those cloud computing, interdependent across locations and, therefore, need coordination. In this paper, laws...

Most of data centers are significantly underutilized. One the major reasons is big gaps between real usage and provisioned resources. In this paper, we first conduct an in-depth analysis a Google cluster trace to unveil root causes for low utilization highlight great potential improve it. We then developed online resource manager Flex maximize while satisfying Quality Service (QoS). Large-scale evaluations show that admits up 1.74× more requests 1.6× higher compared tradition schedulers...

No abstract available.

The power management in micro grid (MG) is related to the planning of dispatchable sources such as diesel generator or storage devices due fluctuation output renewable like wind and solar. aim energy for islanded mode source load shedding. This problem will be considered a multi-objective this paper as: minimize cost electricity production, amount shed day, some specific moments. These objectives are usually conflicted. proposed an approach solving problems, basing on Particle Swarm...

Simultaneously supporting latency- and throughout-sensitive workloads in a shared environment is an increasingly more common challenge big data clusters. Despite many advances, existing cluster schedulers force the same performance goal - fairness most cases on all jobs. Latency-sensitive jobs suffer, while throughput-sensitive ones thrive. Using prioritization does opposite: it opens up path for latency-sensitive to dominate. In this paper, we tackle challenges both short-term long-term...

Data centers are giant factories of Internet data and services. Worldwide consume energy emit emissions more than airline industry. Unfortunately, most significantly underutilized. One the major reasons is big gaps between real usage provisioned resources because users tend to over-estimate their demand center operators often rely on users' requests for resource allocation. In this paper, we first conduct an in-depth analysis a Google cluster trace unveil root causes low utilization...