I love reading articles and projects on the Internet from all fields of programming, computer science, engineering and technology. The problem is that so much information easily wears me out. I usually put the posts in Pocket to read later but my Pocket items just keep accumulating (Yes, I'm a procrastinator). The information is not digested before going out so the good contents on the Web are not made good use of. That has troubled me for long and finally I've made my mind to write a weekly summary of what I've read over the previous week, also in the spirit of others, e.g. Cake Solutions Team Blogs, Fight with Infinity, Web Development Reading List, Hadoop Weekly. Note that it's the time when I read those articles not when they are published. As a analogy, it's processing time rather than event time in a streaming system and lateness is inherent.

Ok, here comes the first one.

Distributed Systems

• Werner Vogels, CTO of Amazon.com, looked back over the past 10 years of AWS and shared with us 10 Lessons from 10 Years of Amazon Web Services.

Streaming

• Apache Flink introduced first version of Complex Event Processing library with the example of monitoring and alert generation for a data center.

• Storm 1.0.0 released with a great many features.

  • up to 16 times faster with latency reduced up to 60%
  • high volume of heartbeat writes to Zookeeper from workers has become a bottleneck for scalability. Now Storm uses Pacemaker, an in-memory key/value store to process heartbeats from workers
  • distributed cache API to allow for sharing of files among topologies with backends of local file system and HDFS
  • Nimbus HA leveraging distributed cache API for replication
  • native Streaming window API
  • stateful Bolt API with automatic checkpointing
  • automatic backpressure
  • resource aware scheduler
  • dynamically change log level for a running topology
  • tuple sampling, distributed log search and worker profiling on Storm UI

• Kafka Streams is a Java library for building distributed stream processing apps using Apache Kafka. To get a sense of it, Introducing Kafka Streams: Stream Processing Made Simple is a nice read from Jay Kreps (author of Kafka, Samza, Confluent co-founder and CEO). I urge you to read his masterpieces if you haven't.

  • The Log: What every software engineer should know about real-time data's unifying abstraction
  • Questioning the Lambda Architecture

• From the industry, Uber has developed a lambda-like architecture for Real-time Stream Computing & Analytics.

Machine Learning

• I've been studying the basics of machine learning from Machine Learning Mastery. For last week,

  • Popular Deep Learning Libraries
  • K-Nearest Neighbors for Machine Learning
  • Naive Bayes for Machine Learning

• Another blog The Beauty of Machine Learning tries to explain machine learning / deep learning in short words.

  • In a Nutshell: Learning
  • In a Nutshell: Neural Networks

• If you've often heard of "Backpropagation" but don't know what it is like me, Calculus on Computational Graphs: Backpropagation is a must read.

• Have you ever wondered how Google search by image works ? Here is the answer, Deep Learning And Machine Learning Simply Explained - Nanalyze

• It's interesting to look at neural networks as functions in functional programming ? Neural Networks, Types, and Functional Programming

• And, Looking inside machine learning black boxes by Julia Evans, whose blog is of great contents.

• Further, explain the predictions of any classifier with LIME - Local Interpretable Model-Agnostic Explanations

• Finally, an interview with DeepMind founder Demis Hassabis, DeepMind founder Demis Hassabis on how AI will shape the future

JVM / GC

• Gearpump recently hit abnormal gc behavior of Executor, where a lot of akka.dispatch.AbstractNodeQueue$Node objects are generated. It turns out to be an Akka issue and GC 'Nepotism' And Linked Queues has given a thorough explanation.

Network / Web

• Once Again on TCP vs UDP weighs up their pros and cons but there is always a trade off.

  In a sense, replacing TCP with UDP is trading off reliability for interactivity. The most critical factor in selection of one over another one is usually related to acceptable delays.

• Security Collapse in the HTTPS Market from ACM Queue assesses legal and technique solutions to secure HTTPS. It's essential to understand the economic incentives of the stakeholders, web-site owners, certificate authorities, web browsers, and end users.

• Google Cloud experienced a connectivity issue in all regions last week. Check Google Cloud Status for detailed description and root cause. TL;DR, it's software bugs.

History

• Judah Levine originally built servers and programmed software to send time over the Internet for U.S. National Institute of Standards and Technology (NIST) back in 1993. Meet the Guy Whose Software Keeps the World’s Clocks in Sync.

Ok, This is the end of first week.