Search

The article discusses SYNERGY, a software-centric approach to field-programmable gate array (FPGA) virtualization in data centers. This approach addresses performance and power challenges by enabling features like task suspension, resumption, migration, and process isolation. Unlike traditional hardware-centric methods, SYNERGY rewrites Verilog hardware description code to work with a Verilog JIT compiler, allowing for higher-level state management and task handling. While it incurs significant overhead and requires further development to match current software virtualization performance, SYNERGY's innovative perspective could lead to improved FPGA architecture and reduced overhead in the future.

The article discusses the significance of a cooling method used in computer data centers known as immersion cooling that could mitigate energy costs and consumption. It examines advances in immersion cooling technology, the advantages of immersion cooling over traditional cooling systems, and the impact upon the measurement rating known as power usage effectiveness (PUE).

The author discusses the issue of power consumption in server farms. She mentions the carbon footprint of some of the biggest companies storing data, the management and distribution of electricity in a data center, and discusses processes for using energy more efficiently.

Planet-scale applications are driving the exponential growth of the Cloud, and datacenter specialization is the key enabler of this trend. GPU- and FPGA-based clouds have already been deployed to accelerate compute-intensive workloads. ASIC-based clouds are a natural evolution as cloud services expand across the planet. ASIC Clouds are purpose-built datacenters comprised of large arrays of ASIC accelerators that optimize the total cost of ownership (TCO) of large, high-volume scale-out computations. On the surface, ASIC Clouds may seem improbable due to high NREs and ASIC inflexibility, but large-scale ASIC Clouds have already been deployed for the Bitcoin cryptocurrency system. This paper distills lessons from these Bitcoin ASIC Clouds and applies them to other large scale workloads such as YouTube-style video-transcoding and Deep Learning, showing superior TCO versus CPU and GPU. It derives Pareto-optimal ASIC Cloud servers based on accelerator properties, by jointly optimizing ASIC architecture, DRAM, motherboard, power delivery, cooling, and operating voltage. Finally, the authors examine the impact of ASIC NRE and when it makes sense to build an ASIC Cloud. [ABSTRACT FROM AUTHOR]

The article discusses the use of Google's tensor processing units (TPUs) to improve performance per watt of deep neural networks in the company's datacenters, with better energy efficiency than central processing units (CPUs) and graphics processing units (GPUs) in similar technologies.

The article summarizes several articles presenting research on the use of distributed cluster scheduling for datacenters. Articles cited include "Large-Scale Cluster Management at Google With Borg" by A. Verma, et al. in the publication "Proceedings of the 10th European Conference on Computer Systems"; "Mesos: A Platform for Fine-Grained Resource Sharing in the Datacenter" by B. Hindman, et al. in the publication "Proceedings of the Usenix Conference on Networked Systems Design and Implementation:; and "Sparrow: Distributed, Low-Latency Scheduling" by K. Ousterhout, et al. in the publication "Proceedings of the Symposium on Operating Systems Principles."

The article discusses techniques for Web-scale computing in relation to autonomous operation of server infrastructure. Topics include the nature of Web-scale datacenters, the nature of operations in relation to server staff, and the nature of software for the stateless servers used as Web-scale. The deployment and testing of software at Web-scale is addressed.

Datacenter workloads demand high computational capabilities, flexibility, power efficiency, and low cost. It is challenging to improve all of these factors simultaneously. To advance datacenter capabilities beyond what commodity server designs can provide, we designed and built a composable, reconfigurable hardware fabric based on field programmable gate arrays (FPGA). Each server in the fabric contains one FPGA, and all FPGAs within a 48-server rack are interconnected over a low-latency, high-bandwidth network. We describe a medium-scale deployment of this fabric on a bed of 1632 servers, and measure its effectiveness in accelerating the ranking component of the Bing web search engine. We describe the requirements and architecture of the system, detail the critical engineering challenges and solutions needed to make the system robust in the presence of failures, and measure the performance, power, and resilience of the system. Under high load, the large-scale reconfigurable fabric improves the ranking throughput of each server by 95% at a desirable latency distribution or reduces tail latency by 29% at a fixed throughput. In other words, the reconfigurable fabric enables the same throughput using only half the number of servers. [ABSTRACT FROM AUTHOR]

In this article, various experts on computer science (CS) offer guides relating to the best of CS research. Particular focus is given to how to determine what papers are worth taking the time to read. Additional topics discussed include the ACM (Association for Computing Machinery), how datacenters are changing the way Web servers are designed and the notion of Network Functions Virtualization (NFV).

The article describes the knowledge gained and lessons from developing and operating three software container management systems, the Borg, Omega, and Kubernetes, all from Internet search company Google. Topics discussed are the transformation of the data center from machine-oriented to being an application-oriented infrastructure, things to avoid when developing the systems such as port numbers management, lack of labeling and ownership uncertainty, as well as application configurations.

The article discusses the implications of high-speed non-volatile storage devices, also known as storage class memories (SCM), for the field of computer science. According to the authors, SCMs invalidate the common assumption that input/output (I/O) devices are slower than central processing units (CPUs). It is suggested that SCMs will also improve the relative performance of layers in systems and make much of the existing enterprise datacenter infrastructure obsolete. Horizontal scaling and workload-aware storage are discussed.

The article explores some of the fundamental issues related to cloud computing. Topics discussed include many components of a cloud computer that systematically deal with the failure of those components, a common architectural pattern used to manage transient failures and the use of simple analytic techniques to allow quantitative statements. Also provided are approaches to addressing frequent server-level failures in a data center, including the decomposition of the system into one or more tiers of servers.

This article describes Google's implementation of a distributed Cron service, serving the vast majority of internal teams that need periodic scheduling of compute jobs. During its existence, we have learned many lessons on how to design and implement what might seem like a basic service. Here, we discuss the problems that distributed Crons face and outline some potential solutions. Cron is a common Unix utility designed to periodically launch arbitrary jobs at user-defined times or intervals. We will first analyze the basic principles of Cron and its most common implementations and then review how an application such as Cron can work in a large, distributed environment, in order to increase the reliability of the system against single-machine failures. We describe a distributed Cron system that is deployed on a small number of machines, but is capable of launching Cron jobs on machines across an entire datacenter, in conjunction with a datacenter scheduling system. [ABSTRACT FROM AUTHOR]

The author discusses advancements in the use of field-programmable gate arrays (FPGAs) in datacenters.

An introduction to an article about application-specific integrated circuit (ASIC) clouds and datacenter specialization, published in the July 2020 edition of the journal, is presented.

The article reports on closing of corporate data centers due to the transition to cloud storage.

The article discusses research on Field Programmable Gate Arrays (FPGAs) in a Microsoft Bing datacenter that enable more efficient search processing at the digital logic level, Topics include Microsoft's production deployment of FPGA processors in its datacenter servers for a range of cloud services, the use of FPGA technology as an alternative to application-specific integrated circuits, and the ability of FPGAs to avoid abstraction overheads and gain unrestricted optimization freedom.