مشخصات مقاله:
عنوان فارسی مقاله:
زیرساخت های محاسباتی مبتنی بر SoC برای برنامه های کاربردی علمی و خدمات تجاری: ارزیابی های اقتصادی و کارایی
عنوان انگلیسی مقاله:
SoC-based computing infrastructures for scientific applications and commercial services: Performance and economic evaluations
کلمات کلیدی مقاله:
سیستمهای مبتنی بر تراشه قدرت پایین، معیار N-body، یادگیری عمیق، توالی نسل بعدی، ارزیابی های اقتصادی و کارایی
مناسب برای رشته های دانشگاهی زیر:
مهندسی کامپیوتر
مناسب برای گرایش های دانشگاهی زیر:
الگوریتم و محاسبات – معماری سیستم های کامپیوتری
وضعیت مقاله انگلیسی و ترجمه:
مقاله انگلیسی را میتوانید به صورت رایگان با فرمت PDF با کلیک بر روی دکمه آبی، دانلود نمایید. برای ثبت سفارش ترجمه نیز روی دکلمه قرمز رنگ کلیک نمایید. سفارش ترجمه نیازمند زمان بوده و ترجمه این مقاله آماده نمیباشد و پس از اتمام ترجمه، فایل ورد تایپ شده قابل دانلود خواهد بود.
فهرست مطالب:
Outline
Highlights
Abstract
Keywords
۱٫ Introduction
۲٫ Materials and methods
۳٫ The N-body application
۴٫ The deep learning application
۵٫ The NGS application
۶٫ Conclusions and future works
Acknowledgments
References
قسمتی از مقاله انگلیسی:
Abstract
Energy consumption represents one of the most relevant issues by now in operating computing infrastructures, from traditional High Performance Computing Centers to Cloud Data Centers. Low power System-on-Chip (SoC) architectures, originally developed in the context of mobile and embedded technologies, are becoming attractive also for scientific and industrial applications given their increasing computing performances, coupled with relatively low costs and power demands. In this paper, we investigate the performance of the most representative SoCs for a computational intensive N-body benchmark, a simple deep learning based application and a real-life application taken from the field of molecular biology. The goal is to assess the trade-off among time-to-solution, energy-to-solution and economical aspects for both scientific and commercial purposes they are able to achieve in comparison to traditional server-grade architectures adopted in present infrastructures.
1. Introduction
Nowadays the costs related to the running of applications, and consequently the provisioning of services, are more and more dominated by the electricity bill, therefore the adoption of energyefficient solutions is crucial [1]. This is particular important for large Information Technology (IT) infrastructures, such as High Performance Computing (HPC) facilities or Cloud Data Centers, which demand energy efficient servers to keep their carbon footprint within acceptable limits. In fact, as reported in [2], data centers in the U.S. consumed in 2014 about 70 billion KWh, representing 1.8% of total U.S. electricity consumption. The perspective is to consume approximately 73 billion KWh in 2020. A similar issue holds true also in the HPC research field, which is facing major challenges on the path towards exascale computing. An exascale system can be defined as a supercomputer that can solve scientific problems 10x faster than the best petaflop systems1 in a power envelope of 20–30 MW. By simply scaling current technologies to exascale, a supercomputer would consume 100 MW, i.e. 10x more than today. Thus, the need for new solutions arises.