diab0lic

26 ms·

271.
by diab0lic 11y ago
You'd be surprised at what a good job the individual pieces of Amazon's empire do at operating as individual units as far as competition is concerned. It would be VERY bad business to "screw" an AWS customer because they
272.
by diab0lic 11y ago
I used to receive multiple push notifications a day at a minimum, which was irritating to me because prior to this a push notification meant that someone was interacting with me, and that didn't hold true anymore. Twitter had broken th
273.
by diab0lic 11y ago
I think this pretty much nails the wall street perspective. Interestingly enough I left the party because their attempts to monetize were really invasive; Emails, many notifications on my phone, etc... I think some subset of their users (to
274.
by diab0lic 11y ago
Cloud Functions / Lambda are great for setting up things such as serverless architecture but they lack a few things necessary for typical "fast data" or stream processing tasks. One could do significantly more with these if t
275.
How to become a Bayesian in eight easy steps(alexanderetz.com)
4 points by diab0lic 11y ago|0 comments
276.
Netflix Tech Blog: Creating Your Own EC2 Spot Market – Part 2(techblog.netflix.com)
3 points by diab0lic 11y ago|0 comments
277.
by diab0lic 11y ago
I don't believe there are any limits on the option program contribution. I'm not trying to refute your point, I imagine we just both have a philosophical difference in view on wether or not this is good.
278.
by diab0lic 11y ago
Right, I apologize, I had assumed you're familiar with the Netflix option program and just inappropriately using the term ESPP. We have a rather straightforward option program in which you purchase an immediately vesting option for 40%
279.
by diab0lic 11y ago
No problem, anything I can do to help you make your decision.
280.
by diab0lic 11y ago
Well I appreciate all the swag and whiteboard markers that come over the wall. :)
281.
by diab0lic 11y ago
Repeated major failures would be something like regularly failing to deliver or making a critical error multiple times. A pattern of behavior that doesn't fit the culture, NOT a one off mistake. A great example of this would be the eng
282.
by diab0lic 11y ago
Thrown work at me? Or are you in the CORE bullpen next to me and have physically thrown things over the wall at me? :P
283.
by diab0lic 11y ago
I currently work at Netflix. Opinions contained herein are entirely my own, sorry for any errors I typed this on a phone. Hope I can help you make a decision. More than willing to answer any questions, just reply here. :) The turnover rate
284.
by diab0lic 11y ago
I didn't consider statistical process control for this task when I did my initial research. It's certainly something I wish I had read about earlier. I have however since started investigating it for a related real-time project th
285.
by diab0lic 11y ago
It's just a matter of collecting and tagging enough. We've been working on it with our new operational data tagger. Keep an eye out for a future post! :)
286.
by diab0lic 11y ago
One of my colleagues designed and implemented a telemetry data tagger that allows service owners to annotate the data reported into our primary telemetry system Atlas [0] and has further worked on mechanisms for service owner feedback to au
287.
by diab0lic 11y ago
Hey there, We have considered modeling the distribution from which the data is typically drawn and then calculating likelihood of newly observed data. Some of the approaches that we use to detect anomalies on stream starts per second (SPS)
288.
by diab0lic 11y ago
We most certainly do, one of the more obscure applications is detecting customer service agents who are gaming the system.
289.
by diab0lic 11y ago
We have, specifically in the domain of using RL for parameter selection. Part of the reason I was brought on board was background experience in RL. I fully believe it's the key to turning domain (yes/no) decisions into numbers for
290.
by diab0lic 11y ago
This was something I looked into when i performed the initial investigation for this project. It was a bit difficult to locate supporting academic material on the algorithm though. The white paper on the page seemed more like marketing mate
291.
by diab0lic 11y ago
Our central alerting gateway (CAG) does much more than just send emails and pages. When the system was fist built we hooked into CAG directly as it is capable of taking many of the desired actions on its own.
292.
by diab0lic 11y ago
No problem, thanks for taking the time to read it! :D
293.
by diab0lic 11y ago
We have another component of the same outlier detection system that does this type of fitting, identifying low probability events using a Bayesian model + Markov Chain Monte Carlo. It hasn't gained nearly as much traction internally (y
294.
by diab0lic 11y ago
It would certainly help separability to observe the data in a higher dimensional space, however when you're taking automated action against the results its sometimes pertinent to know which specific metric is causing the server to be a
295.
by diab0lic 11y ago
Right you are!
296.
by diab0lic 11y ago
We tend to operate on sever performance metrics, error rates, and networking metrics. We've found through practice that these metrics tend to reveal most issues that we're targeting. If you meant in the more mathematical sense we
297.
by diab0lic 11y ago
Thanks so much dang!
298.
by diab0lic 11y ago
One of the more complex situations that this solution attempts to address is the situation where the entire cluster surges upwards. Imagine that service A depends on service B, and service B begins experiencing an issue. The outlier detecti
299.
by diab0lic 11y ago
RAD requires at least 2 cycles of data, whereas the real-time analytics team tends to operate on very short time frames. Our approach is less accurate but able to make quick decisions, which is the objective in this case. Both systems are c
300.
by diab0lic 11y ago
The answer is most certainly "both". The system can be used to identify problems you don't know you had (generally sending an email / page). Once you've identified a problem you may have it pull instances out of our

More ›