Explore chapters and articles related to this topic
Hybrid Energy Systems for Computing and Electronic Industries
Published in Yatish T. Shah, Hybrid Energy Systems, 2021
The comparison showed that there was no clear winner in terms of performance per watt. Depending on the workload, different platforms showed best performance per watt (i.e., power efficiency). For data mining workloads (I/O bound), both low power architectures showed a clear advantage (Atom 330 is 3-4x better than Xeon), while for more traditional web or compute-intensive workloads, the Xeon server was still the platform of choice. This suggests that mixing different platforms can be better in terms of power efficiency with diverse workloads. Further, the least power-hungry architectures, the Atom N270, exhibited the best performance/watt for Word Count compared to the other servers, but very little gain in performance/watt compared to Xeon processors for other workloads. This is due to the specific mini-PCI (peripheral component interconnect) solid-state drive used in the system that provides good read throughput but very low write throughput. Word Count benefits from this characteristic as it is mainly reads.
Modular Systems for Energy Conservation and Efficiency
Published in Yatish T. Shah, Modular Systems for Energy Usage Management, 2020
Due to the rising power costs, energy efficiency is growing as a top concern of data center managers. DellTM PowerEdgeTM M1000e infrastructure directly improves the energy efficiency of current and future data centers through a combination of hardware and software improvements. Through higher power density, load balancing, and improved energy efficiency, hardware directly affects a data center’s operating cost. The Dell Chassis Management Controller (CMC), which is a critical part of all Dell blade servers, provides software features that help the 2,700 watts (W) power supply perform at maximum efficiency. Features dynamic power supply engagement (DPSE), maximum power conservation mode, and power monitoring help data center managers push the limits of the M1000e modular blade server enclosure. With data centers increasing in density and requiring more performance per watt, more power at higher efficiency is required. The Dell PowerEdge M1000e infrastructure meets the demand of the modern data center and pushes power supply technology to the cutting edge. Higher alternating current-direct current (AC-DC) conversion efficiency, extreme capacity, and innovative algorithms have ushered the 2,700 W power supply to the next level of modular infrastructure. The AC-DC conversion has reached a new high of 94% efficiency, with an increase of up to 13% in power output over prior consumption. When these modular hardware advantages are coupled with the power features in the Dell ChCMC, the amount in power efficiency is very high.
Energy-efficient distributed password hash computation on heterogeneous embedded system
Published in Automatika, 2022
Branimir Pervan, Josip Knezović, Emanuel Guberović
Bcrypt implementations on homogeneous and heterogeneous multiprocessing platforms: Parallella board with 16- or 64-core Epiphany accelerator and ZedBoard showed better performance per Watt compared to CPU implementations [31]. These implementations were integrated into John the Ripper password cracker resulting in improved energy efficiency by a factor of 35+ compared to heavily optimized implementations on modern CPUs.
A green proactive orchestration architecture for cloud resources
Published in International Journal of Computers and Applications, 2019
Our hosting cloud model represents the underlying infrastructure as a large-scale data center hosting various web applications. The assumed data center comprises N heterogeneous physical servers, where physical resources are shared among the hosted applications in a virtualization environment. Similar to [24,25], in order to assess the performance of the employed servers, the Performance Per Watt (PPW) metric is used. PPW is a measure of energy efficiency of a particular architecture or computer hardware [26]. Here, PPW can be defined as the rate of transactions or computations that can be delivered by a computer for every watt of power consumed [25]. Each physical server offers a pool of virtual computing units (VCUs). Similar systems have been investigated, for example [22,23,27,28]. Each hosted application has a specific SLA defining the QoS that the application is anticipating to grantee from the system. Clients send demanding requests to be processed by the host web application which will arrive to the adopted model according to a stochastic process that may change over time. Each hosted web application expresses a demand for provisioning a dedicated VM. Each VM with a defined priority derived from the criticality of the application itself, definite resource requirements expressed as Millions of Instructions Per Second (MIPS), exact starting and finishing times, a specific amount of RAM, along with never/always together VMs list. In order to determine workload requirements, this work assumed that a priori profiling stage is employed, where VMs are firstly profiled on a staging server to determine the amount of physical resources needed to achieve the negotiated SLA-level. Also, during this stage, the submitted VM is assigned a predefined priority that comes from how critical is this web service application. For example, A VM belonging to critical business applications takes higher priority than a VM belonging to ordinary web applications. It is also assumed that each client request can be treated independently of other requests. Each request has an SLA derived from the QoS that the targeted application should guarantee from the cloud system. In this work, the SLAs contain mainly the average response time for each request, meaning that an application should have a specific share of the CPU capacity, thus their clients experience an acceptable response time from the cloud system. An SLA violation occurs when a VM cannot get the requested amount of resources.