The tradeoffs of large scale learning

The tradeoffs of large scale learning Bottou & Bousquet, NIPS’07

Welcome to another year of The Morning Paper. As usual we’ll be looking at a broad cross-section of computer science research (I have over 40 conferences/workshops on my list to keep an eye on as a start!). I’ve no idea yet what papers we’ll stumble across, but if previous years are anything to go by I’m sure there’ll be plenty of great material to keep interest levels high.

To start us off, today’s paper choice is “The tradeoffs of large scale learning,” which won the ‘test of time’ award at NeurIPS last month.

this seminal work investigated the interplay between data and computation in ML, showing that if one is limited by computing power but can make use of a large dataset, it is more efficient to perform a small amount of computation on many individual training examples rather than to perform extensive computation on a subset of the data. [Google AI blog: The NeurIPS 2018 Test of Time Award].

For a given time/computation budget, are we better off performing a computationally cheaper (e.g., approximate) computation over lots of data, or a more accurate computation Continue reading

StackStorm ChatOps Integration

StackStorm is IFTTT for ops. Its original claim to fame was popularizing chat driven workflows for ops know as ChatOps. In this post I will cover integrating ST2 with Slack to enable ChatOps. Lab Environment The ST2 instance connects out to Slack which is a SAAS application. This means...

oVirt and OKD

This is a series of posts to demonstrate how to install OKD 3.11 on oVirt and what you can do with it. Part I - How to install OKD 3.11 on oVirt

How to install OKD 3.11 on oVirt (4.2 and up)

Installing OKD or Kubernetes on oVirt has many advantages, and it's also gotten a lot easier these days. Admins and users who want to take container platform management for a spin, on oVirt, will be encouraged by this.
Few of the advantages are:

  • Virtualizing the control plane for Kubernetes - provide HA/backup/affinity capabilities to the controllers and allowing hardware maintenance cycles
  • Providing persistent volume for containers via the IAAS, without the need for additional storage array dedicated to Kubernetes
  • Allowing a quick method to build up/tear down Kubernetes clusters, providing hard tenancy model via VMs between clusters.

The installation uses openshift-ansible and, specifically the openshift_ovirt ansible-role. The integration between OpenShift and oVirt is tighter, and provides storage integration. If you need persistent volumes for your containers you can get that directly from oVirt using ovirt-volume-provisioner and ovirt-flexvolume-driver.
For the sake of simplicity, this example will cover an all-in-one OpenShift cluster, on a single VM.
Continue reading

How DevOps Should Use DBaaS (Database-as-a-Service) To Optimize Their Application Development

This post was written by Wendy Dessler of The Blog Frog.

Database-as-a-Service (DBaaS) is quickly gaining in popularity across the tech world. These software platform solutions helps users easily manage their database operations without having to really understand any of the abstractions. This allows developers, DBA’s and DevOps engineers to quickly automate their backups, create new SQL and NoSQL clusters, and monitor the performance of their databases for their application without requiring any internal database expertise.

DBaaS falls under the umbrella of Platform-as-a-Service (PaaS) where the platform itself is actually a database or several databases. This is a great choice for DevOps in particular because it allows for more developer agility, productivity, and also security.

Flexibility and scalability are becoming more important in the world of DevOps and technology in general, and we all know how fast this world moves. Businesses need new ways to keep up with the competition, and developers are looking for an easy, self-service model for managing their databases in order to optimize their app development. Let’s break down the individual benefits so you can decide if DBaaS is right for your DevOps team.

1. Outsourced Security and Administration

Whither Network Engineering? (Part 3)

In the previous two parts of this series, I have looked at the reasons I think the networking ecosystem is bound to change and why I think disaggregation is going to play a major role in that change. If I am right about the changes happening, what will become of network engineers? The bifurcation of knowledge, combined with the kinds of networks and companies noted in the previous posts in this series, point the way. There will, I think, be three distinct careers where the current “network engineer” currently exists on the operational side:

  1. Moving up the stack, towards business, the more management role. This will be captured primarily by the companies that operate in market verticals deep and narrow enough to survive without a strong focus on data, and hence can survive a transition to black box, fully integrated solutions. This position will largely be focused on deploying, integrating, and automating vertically integrated, vendor-driven systems and managing vendor relationships.
  2. Moving up the stack, towards software and business, the disaggregated network engineering role (I don’t have a better name for this presently). This will be in support of companies that value data to the point of focusing on its management Continue reading

Stuff The Internet Says On Scalability For January 4th, 2019

Wake up! It's HighScalability time:

 

Solar system? Nope, the beauty is in your head—neural art.

 

Do you like this sort of Stuff? Please support me on Patreon. Need cloud? Explain the Cloud Like I'm 10 (34 almost 5 star reviews).

 

  • 45%: learned scheduler improves average job completion time; 61%: apps share data with Facebook; 45,037,125: people who watched Bird Box on Netflix in first week; 32,368: color images collected by the Curiosity rover on Mars between August 2012 and November 2018; $36.1B: AI  healthcare market by 2025; 20%: object recognition failure in light rain; 350: pages in Donald Knuth's new book;  $1,000: price needed to deactivate Facebook account; $3B: Epic games profit; 1: bitcoin mined from body heat of 44,000 people; 

  • Quotable Quotes:
    • @Hannah_Chutzpah: What are the technical terms, in your field, for 'dunno'? In medicine there's 'idoeopathic' In archeology/anthropology there's 'ritual purposes' How do you professionally term 'we haven't got a clue'?
      • @peterseibel: In programming we don't have a special term for it, just like fish don't have a special term for 'water'.
    • @robn: everyone seems to want stateless services and stateless protocols but Continue reading
1 … 1,410 1,411 1,412 1,413 1,414 … 3,895