Skip to content

Latest commit

 

History

History
154 lines (118 loc) · 7.03 KB

2017-W34.md

File metadata and controls

154 lines (118 loc) · 7.03 KB

Works on ARM news, Issue 17, August 25, 2017 (Week 34)

Works on ARM news is written and edited by Ed Vielmetti at Packet. It comes out weekly on Fridays on Twitter's @worksonarm account and via Github and Gitbook. Subscription details are at the bottom of this issue.

In this issue

  • Qualcomm Centriq 2400 architecture, first view
  • Edge Computing (Wired)
  • Your Serverless Raspberry Pi cluster with Docker (Alex Ellis)
  • Towards deploying decommissioned mobile devices as cheap energy-efficient compute nodes
  • ODroid HC-1 Home Cloud (ARMv7)

Qualcomm Centriq 2400 architecture, first view

Qualcomm has released more information about their Centriq 2400 architecture to the public via a presentation at the Hot Chips conference. The electronics technology press has run a series of stories linked below about what's new and notable in these announcements.

Falkor is the custom CPU design at the heart of the Qualcomm Centriq 2400 SoC, the world’s first 10nm server processor, which will begin shipping commercially later this year.

The press release and presentations focus on several features of the Centriq 2400. The system is a pure 64-bit ARMv8 compliant architecture, with no 32-bit cores in the design, and draws from Qualcomm's history of developing power-efficient custom mobile processors. The 48-core design uses 24 pairs of custom Falkor CPUs which share an L2 cache and bus interconnect to the high speed Qualcomm System Bus (QSB). Power efficiency is engineered into the hardware, with fast recovery from sleep for idle cores.

Falkor supports cryptographic instructions in the ARMv8 instruction set for performance when running networking security protocols such as https. The system is delivered on a fully-integrated SoC with a high-bandwidth, low latency ring interconnect to a L3 cache.

Much more information is in the press kit and in the detailed walk-through published by Anandtech. Expect more announcements before the launch. Ian Cutress has a particularly detailed 7-part analysis published on Anandtech, which is a good read.

Edge Computing (Wired)

In "It's time to think beyond cloud computing", Jeremy Hsu writes for Wired Magazine's "Backchannel" column about how computing workloads will migrate to the edge of the network and away from monolithic centralized data centers. He quotes Packet's CEO Zac Smith:

It’s a foregone conclusion that giant, centralized server farms that take up 19 city blocks of power are just not going to work everywhere.

Packet has just announced an expansion to six additional US data centers and five additional international data centers. Also mentioned in the article is Austin-based Vapor IO's buildout of micro data centers in cell tower infrastructure.

The interesting point mentioned here is that the Internet did not always have centralized data centers in its architecture. Peter Levine's presentation on "The End of Cloud Computing" from Andreessen Horowitz in 2016 notes that there are cycles in computing between centralized and decentralized, and that demands from latency-sensitive applications (like real-time information for self-driving cars) will push more compute away from centralized data farms into distributed systems.

No doubt that power-efficient ARM-based designs will find new uses and new users at the edge of the network.

Your Serverless Raspberry Pi cluster with Docker (Alex Ellis)

Alex Ellis has a fine writeup on the hardware behid a "serverless" architecture on the ARMv7 based Raspberry Pi and his OpenFaaS functions-as-a-service architecture. He notes:

This blog post will show you how to create your own Serverless Raspberry Pi cluster with Docker and the OpenFaaS framework. People often ask me what they should do with their cluster and this application is perfect for the credit-card sized device - want more compute power? Scale by adding more RPis.

Alex writes that "serverless" is a metaphor for this computing style, where instead of building out a system out of virtual machines, you reduce the workload to the exercise of individual functions.

Towards deploying decommissioned mobile devices as cheap energy-efficient compute nodes

Shahrad & Wentzlaff at HotCloud’17 have a presentation and paper on using old mobile phones as energy-efficient compute nodes. They point out that innovation in the cell phone market has led to ever more efficient and powerful devices even at the low end, and that clever power and housing designs could readily allow a lab environment to build a highly capable system out of Android-powered cell devices racked into a 2U chassis and powered and networked using USB.

Adrian Colyer writes of this work:

I have one simple rule when it comes to selecting papers for The Morning Paper: I only cover papers that I like and find interesting. There are some papers though, that manage to generate in me a genuine feeling of excitement, as in “this is so cool, I can’t wait to share it.” This is one of those papers!

ODroid HC-1 Home Cloud (ARMv7)

ODroid and Hardkernel have announced a new ODroid HC-1 "Home Cloud" ARMv7 32-bit device which is specifically designed for attaching to an SSD or HDD for network-attached storage. The device design is a cost reduction and simplification of the ODroid XU-4, based on the Samsung Exynos 5422 octa-core Cortex-A15/A7 processor. The nice industrial design features an aluminum case that doubles as a heat sink for the SoC and is easily stackable.

Subscription information

Works on ARM News is released weekly on Fridays. Follow @worksonarm on Twitter for the weekly link, or follow the worksonarm-news project on Github for notifications. Email to [email protected] if you'd like a direct email.

Back and current issues are available from the Works on ARM News archive on Github.