Posts

Showing posts from 2016

The Curious Case of vCloud Director and the Perils of Over-Engineering

Back in December 2011 , I wrote what many considered a borderline heretical blog post: a prediction that VMware vCloud Director,  that shining paragon of private cloud orchestration, would not thrive in the following five years. At the time, I was accused of being overly skeptical, anti-vCloud, and even of being secretly on AWS’s payroll. (Spoiler: I wasn’t. Though I wouldn’t have minded the stock options.) Now that it’s August 2016 , I figured it’s only fair to revisit that time capsule of cynicism and ask: Was I right?  Well, grab your favorite performance dashboard, because the data doesn’t lie. Let’s tally up the results. What I Warned in 2011 Just to refresh your memory (and my ego), here’s the TL;DR of what I said five years ago: vCloud Director was too complex for most enterprises. It was designed more for telcos and service providers , not your average enterprise IT shop. It introduced too many layers of abstraction , distancing users from the physical in...

Do Neural Networks Dream of Strictly Convex Sheep?

Image
  A parting thought before leaving Herndon, written on a flight to Dallas after a week at Amazon Machine Learning University. About what we wanted from optimization landscapes and what we got instead. Courtesy: Amazon MLU There is a moment in training a deep network on a g2.2xlarge when the loss does something I can only call insolent. It drops for a few epochs, plateaus for what feels like forever, jitters sideways, then drops again into a region where the gradient is essentially noise around a slow downward drift. My textbooks call this descending. My eyes call it wandering. The optimizer is moving through a country we do not have good language for, because the language we have was built for a country that does not exist. That country is convex. I am writing this somewhere over Tennessee after a week in a windowless room learning what Amazon thinks its engineers should know about machine learning. The instructors were good. The math was rigorous. But there was a recurring tension...