Talk 0: Meetup and Technology Updates (Chris Fregly, Research Scientist @PipelineIO)
• We hit 5000 Members!!!!
• Recent Advancements in Spark ML + Tensorflow AI Training and Serving/Predicting (YouTube Video)
• Tensorflow v0.12: HDFS, lots of deprecated APIs ahead of v1.0
• Tensorflow v1.0 coming soon (Tensorflow Dev Summit in February)
• Upcoming O’Reilly Training: High-Performance Tensorflow Model Serving (Chris Fregly, Research Scientist @ PipelineIO)
• Next 2 Meetups (Feb and Mar): Lots of Streaming!!!
Talk 1: Training and Serving Tensorflow AI Models at Scale using Kubernetes, Distributed Tensorflow, Tensorflow Serving, JavaCPP, and NetflixOSS (Chris Fregly, Research Scientist @ PipelineIO)
• Demos, Docker Images, and Open Source Code!
Talk 2: Low-Level CPU Performance Profiling Examples using Apache Spark, Apache Arrow, and Columnar Databases (Tanel Poder, Founder and Chief Engineer @ Gluent)
• In this session Tanel Poder will demonstrate some low level performance tools like Linux “perf stat” to measure the memory access traffic and CPU efficiency of different workloads, data structures and programming paradigms.
• We will use a columnar database, a few variations of a Spark job and Apache Arrow data structure iteration as examples.
• This session’s goals are to emphasize the importance of using suitable data structures for a task (like a columnar structure for scanning) and that modern CPUs and performance tools give you good visibility into the “CPU-friendliness” of your code.
• Tanel Poder is a co-founder of his current startup Gluent that liberates enterprise data, making it useful across all enterprise.
• Despite holding a CEO title, he was an advanced OS & database systems performance geek for over 20 years and is now hoping to bring some of that skill to the Spark/Big Data world too.
Talk 3: Timothy Chen (Former Spark Engineer @ Mesosphere)
1) Deep Dive into Spark + Mesos Integration
• highlight any hot spots with the Spark+Mesos integration
• things Tim struggled with during development
• cool demos of Spark + Mesos
2) Active Work on Spark + Kubernetes Integration
• highlight the latest design proposals and/or working code
• cool demos of Spark + Kubernetes