We're happy to announce the 4th Haifa Cloud Meetup event!

https://secure.meetupstatic.com/photos/event/a/8/0/d/600_464083021.jpeg

I -

Title:

High performance data analytics with the Stocator object storage connector for Apache Spark

Abstract:

With data growing voluminously at high velocity there comes a need for inexpensive and secure way to store and access data. Object store technology, such as IBM Cloud Object Storage, enable big data analytics engine such as Apache Spark to cost effectively analyze and drive value from vast quantities of data. In this talk we will review the differences between file store and object stores and introduce Stocator, a new connector between Apache Spark and object stores which leverages object store semantics to achieve high performance while maintaining fault tolerant and speculative execution. We will deep dive into the low level details of how Apache Spark communicates with an object store, compare different connectors such as s3a, stocator, and hadoop swift, and present best practices for achieving high performance.

Bio:

Effi Ofer (https://www.linkedin.com/in/effi-ofer-91a261b0/)is a researcher at IBM Research at Haifa, currently researching data analytics and cloud object stores.

II-

Title:

Efficiency of GPU for Deep Learning Inference Efficiency of GPU for Deep Learning Inference

Abstract:

After a neural a network is trained we get to inference stage: Doing actual predictions. In many cases inference is slow even when running on GPU, which makes application to big data sets problematic.We performed a research on efficiency of GPU on inference process, in order to identify a bottleneck.Our findings were that in feed forward neural networks with dense layers - memory bandwidth is a bottleneck. In the presentation I am going to elaborate about the research process.

Bio:

David Gruzman is leading startup www.nestlogic.com working on automatic finding insights in big data sets. His professional interests are efficiency of big data processing, usage of ML and DL to unsupervised insights discovery in data.