Revisiting Dynamic Thermal Management Exploiting Inverse Thermal Dependence

Reducing energy consumption without affecting computational performance is a significant research driver in computer engineering. The Partitioned Global Address Space (PGAS) programming model provides a global address space for ease-of-use while providing locality-awareness for efficient execution. For symmetric multiprocessor (SMP) clusters, PGAS locality-awareness offers opportunities for intelligent energy management due to the inherent latencies in the interconnection network fabric. In this paper, the second in a series that explores PGAS power optimization, we present a detailed examination of how data locality-awareness can be exploited to improve the energy consumption of workload sharing on SMP clusters. We present a systematic methodology for analyzing a given algorithm's potential for locality-aware power optimization. We show how to automate the code modifications for an iterative stencil loop algorithm using a profiler that analyzes the algorithms' non-local access patterns and optimizes the code for thread-local data pre-fetch. We explore the use of dynamic voltage frequency scaling (DVFS) in conjunction with time-based optimizations and provide an extensible and precise power measurement framework for generating the experimental results. We compare optimized and CPU-default approaches to DVFS application at the program-execution and processor architecture level. We show how the latency of the cluster's interconnection network can impact the opportunities for power optimization by comparing the programs' energy/time behavior when the cluster nodes are connected by Ethernet or Infiniband network fabrics. For our representative iterative stencil loop algorithm, locality aware power optimizations show significant improvements in energy efficiency over native code running in default processor power modes.

in increased on-chip temperature leading to serious concerns about lifetime reliability of micro-processors. Though dynamic thermal management techniques can control peak temperature, they often fail to meet the reliability targets due to the complex interplay between temperature and reliability. In this paper, we propose a dynamic reliability management (DRM) technique that exploits architectural adaptation in conjunction with dynamic voltage/frequency scaling (DVFS). We employ an online Bayesian classifier that can efficiently detect the reliable configurations, while a performance prediction model selects the one with best performance among all the reliable configurations. We later extend our approach to meet both reliability and thermal constraints. Experimental results reveal that our adaptive DRM technique achieves reliability targets while reducing performance overhead by 42.30 % compared to DVFS and 28.68 % compared to DVFS with fetch gating. I.

Modern information and communication technology along with the broad adaptation of computing paradigms, namely cloud computing has significant impact on environment due to huge amount of CO 2 emission. In the recent decade, a significant growth in the research domain of “green” and low power consumption network has been observed. With the recent advancements of compute intensive machine learning frameworks like deep learning, the energy consumption by GPUs is a challenging issue. In all of these cases, Dynamic Voltage Frequency Scaling (DVFS) has gained significant research interests and shown encouraging outcome. Dynamic Voltage and Frequency Scaling (DVFS) is utilized to reduce the energy consumption in data centers without impacting the Quality of Service. On the other side, spatio-temporal dataset is huge in size and requires compute and time intensive algorithms and techniques for different spatio-temporal query processing. In all aspects, the correlations among energy, DVFS, consolidation, and performance are the key-enablers of energy-efficient management. In this chapter, we explore different existing DVFS techniques and challenges to analyse and process spatio-temporal query in cloud based paradigm. We have considered movement datasets for demonstrating energy consumption and utility of dynamic voltage and frequency scaling.KeywordsDynamic voltage and frequency scalingSpatio-temporal QueryCloud datacenterScheduling algorithmPowerEnergy

Small size of the sensor nodes put constraint on the hardware capabilities and battery supported on it. Though power optimization is a real challenge in all WSN applications, power saving at the cost of data loss is not at all preferable. Very small buffer sizes cannot store the increased data arrival at the time of event occurrences and data gets lost due to buffer overflow. Data loss within the node occurs before transmission mainly either data gets timed out or due to buffer overflows. Both of these situations can be handled using CAP management techniques at individual sensor nodes. Here Dynamic Voltage Frequency Scaling (DVFS) and Dynamic Modulation Scaling (DMS) techniques are used for optimization of power consumed by processor and transmitter respectively. We have considered that DVFS and DMS are coordinated together so that whenever processor changes its processing speed, transmitter also changes its transmission rate and minimum data loss occurs at the intermediate buffer. In this paper we have discussed adaptive threshold policy for selecting a particular state of sensor node and compared the results with fixed threshold policy. CAP management along with adaptive threshold policy improves the lifetime of sensor nodes and also reduces the data loss before transmission.

Employing hardware accelerators to improve the performance and energy-efficiency of DNN applications is on the rise. One challenge of using hardware accelerators, including the GPU-based ones, is that their performance is limited by internal and external factors, such as power caps. A common approach to meet the power cap constraint is using the Dynamic Voltage Frequency Scaling (DVFS) technique. However, the functionally of this technique is limited and platform-dependent. To tackle this challenge, we propose a new control knob, which is the size of input batches fed to the GPU accelerator in DNN inference applications. We first evaluate the impact of batch size on power consumption and performance of DNN inference. Then, we introduce the design and implementation of a fast and lightweight runtime system, called BatchDVFS. Dynamic batching is implemented in <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">BatchDVFS</i> to adaptively change the batch size, and hence, trade-off throughput with power consumption. It employs an approach based on binary search to find the proper batch size within a short period of time. Combining dynamic batching with the DVFS technique, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">BatchDVFS</i> can control the power consumption in wider ranges, and hence, yield higher throughput in the presence of power caps. To find near-optimal solution for long-running jobs that can afford a relatively significant profiling overhead, compared with <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">BatchDVFS</i> overhead, we also design an approach, called BOBD, that employs Bayesian Optimization to wisely explore the vast state space resulted by combination of the batch size and DVFS solutions. Conducting several experiments using a modern GPU and several DNN models and input datasets, we show that our <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">BatchDVFS</i> can significantly surpass the techniques solely based on DVFS or batching, regarding throughput (up to 11.2x and 2.2x, respectively), while successfully meeting the power cap.

DVFS (Dynamic voltage frequency scaling) is one of the most widely used power management technologies employed to improve the performance or minimize the power consumption by controlling voltages and frequencies in real time. When applying device-level DVFS in graphics processing units (GPUs) that support spatial multitasking, it is difficult to determine the optimal DVFS status when multiple running kernels have different characteristics. To solve the problem, we created a GPU simulator that can operate at different streaming multiprocessor frequencies according to the characteristics of the assigned kernel and compared it with a single-clock-based spatial multitasking GPU simulator.

A smart energy and reliability aware scheduling algorithm for workflow execution in DVFS-enabled cloud environment

Multi-core processors have become an integral part of mainstream high performance computer systems. In parallel, exponentially increasing power density and packaging costs have necessitated system level thermal management solutions for multi-core systems. Dynamic thermal management (DTM) techniques monitor on-chip temperature continuously and typically employs dynamic voltage and frequency scaling (DVFS) to lower the temperature when it exceeds a pre-defined threshold. State-of-the-art DTM solutions for multi-core systems include distributed DVFS (where each core can scale the voltage/frequency individually) and global DVFS (where all cores scale voltage/frequency simultaneously). Distributed DVFS generally offers higher performance than global DVFS, but it is hard to implement and has major scalability issues.

This paper proposes to take the relationship between delay and workload into account in the power optimization of microprocessors in mobile embedded systems. Since the components outside a device continuously change their values or properties, the workload to be handled by the systems becomes dynamic and variable. This variable workload is formulated as a staircase function of the delay taken at the previous iteration in this paper and applied to the power optimization of DVFS (dynamic voltage-frequency scaling). In doing so, a graph representation of all possible workload/mode changes during the lifetime of a device, Workload Transition Graph (WTG), is proposed. Then, the power optimization problem is transformed into finding a cycle (closed walk) in WTG which minimizes the average power consumption over it. Out of the obtained optimal cycle of WTG, one can derive the optimal power management policy of the target device. It is shown that the proposed policy is valid for both continuous and discrete DVFS models. The effectiveness of the proposed power optimization policy is demonstrated with the simulation results of synthetic and real-life examples.

The research on green cloud computing is indispensable to reduce the emission of carbon as well as to rescue the environment from the hands of global warming, as the climatic change due to the emission of carbon is a critical issue now a days. The green cloud computing mainly concentrates on reducing the amount of cloud resources to condense the emission of carbon. So an effective and efficient scheduling algorithm, Green Task Scheduling (GTS) Algorithm, is proposed in this paper to bring down the amount of cloud resources in use. The proposed algorithm also reduces the hardware cost significantly. As green cloud computing also concentrates to save the energy consumption by providing only the required amount of voltage based on the frequency of work, a technique called as Dynamic Voltage Frequency Scaling (DVFS) is also used in this paper to control the voltage as well as the frequency of the processor without affecting the working performance. The numerical results prove that the makes pan, energy and cost are optimally minimized by implementing GTS along with DVFS in cloud computing environment.

Nowadays, the expanding computational capabilities of the cloud system rely on the minimisation of the consumed power to make them sustainable and economically productive. Power management of cloud data centres received a great attention from industry and academia as it consumes high energy and thus increases the operational cost. One of the core approaches for the conservation of energy in the cloud data centre is the task scheduling. This task allocation in a heterogeneous environment is a well known NP-hard problem due to which researchers pay attention for proposing various heuristic techniques for the problem. In this paper, a technique is proposed based on dynamic voltage frequency scaling (DVFS) for optimising the energy consumption in the cloud environment. The basic idea is to address the trade-off between energy consumption and makespan of the system. Here, we formally introduce a model that includes various subsystems and assess the implementation of the algorithm in the heterogeneous environment.

With the development of streaming media, soft real-time system in today’s life could participate in the use of more extensive areas. The use frequency was also increasing. Consequently, modern processors were equipped with software control mechanisms such as DVFS (Dynamic Voltage Frequency Scaling) to allow operating systems to meet required performance while reducing power consumption. Therefore, we propose a task scheduling algorithm combined DVFS technology and time deterministic cyclic scheduling to achieve energy saving effect. First, the algorithm needed to minimize the preemption between tasks to reduce latency, so we created a buffer to save periodic tasks to avoid preemption. Second, to reduce the computational cost of the scheduling scheme, a scheduling template were designed to perform tasks. In this paper, the scheduling of periodic tasks, task scheduling would be designed when the task scheduling template would be fixed length. Besides, this algorithm supported that task could adopt appropriate voltage and frequency through DVFS technology in idle time under the condition of satisfying task dependence. Experimental analysis showed that the proposed algorithm could effectively reduce the system energy consumption while ensuring the completion of the task.

Cloud computing is the combination of grid computing, distributed computing and utility computing. Cloud computing provides various types of services (servers and storage facility) in the on demand basis. The main goal of the cloud computing is to preserve and organize the very huge data center or data forms. The data forms are composed of thousands of servers that absorb the giant (ample) amount of electricity in the word of energy. The decreasing the energy consumption in datacenter is the major challenge in cloud computing now a day. This research article is going to address the problem of high energy consumption at datacenter. Concentrate on virtual machine scheduling in cloud datacenter with Dynamic Voltage Frequency Scaling (DVFS) approach. We have combined shortest job first and Round Robin algorithms with Vibrant Quantum. This combination of algorithm is considered as shortest round vibrant queue (SRVQ) algorithm. SRVQ reduces the waiting time of the scheduling process and minimize the starvation. The DVFS and SRVQ worked together and produced fruitful results in the final experiments. This work reduced the server’s energy consumption in the cloud data center. In the final results, our proposed framework exhibits 45% of energy efficiency compare to other previously proposed algorithms. 33% of QoS performance were enhanced by our framework.

We consider the problem of energy-efficient scheduling for slice-parallel video decoders on multicore systems with Dynamic Voltage Frequency Scaling (DVFS) enabled processors. We rigorously formulate the problem as a Markov decision process (MDP), which simultaneously considers the on-line scheduling and per-core DVFS capabilities; the power consumption of the processor cores and caches; and the loss tolerant and dynamic nature of the video decoder. The objective is to minimize long-term power consumption subject to a minimum Quality of Service (QoS) constraint related to the decoder’s throughput. We evaluate the proposed scheduling algorithm using traces generated from a cycle-accurate multiprocessor ARM simulator.

Emerging trends in applications with the requirement of considerable computational performance and decreasing time-to-market have urged the need of multiprocessor systems. With the increase in number of processor there is an increased demand to efficiently control the energy and power budget of such embedded systems as well. Dynamic voltage frequency scaling (DVFS) strategies permits to control this budget by actively changing the power consumption profile of the system. These techniques exploit the execution times variation of Real-time applications for dynamically adjusting the voltage and frequency of processors in order to reduce power and energy consumption. This paper presents one DVFS strategy called Dynamic slack reclamation (DSR) for real time applications. It is based on detecting early completion of tasks. DSR determines the amount of dynamic slack (if any) by comparing the worst-case execution requirement of a tasks job with its actual execution requirements. Experimental results will be extracted from the simulation of MPSoC architectures using the SoCLib platform; they show that DSR approach gives better power consumption performance.