Book Image

Real-Time Big Data Analytics

By : Sumit Gupta, Shilpi Saxena
Book Image

Real-Time Big Data Analytics

By: Sumit Gupta, Shilpi Saxena

Overview of this book

Enterprise has been striving hard to deal with the challenges of data arriving in real time or near real time. Although there are technologies such as Storm and Spark (and many more) that solve the challenges of real-time data, using the appropriate technology/framework for the right business use case is the key to success. This book provides you with the skills required to quickly design, implement and deploy your real-time analytics using real-world examples of big data use cases. From the beginning of the book, we will cover the basics of varied real-time data processing frameworks and technologies. We will discuss and explain the differences between batch and real-time processing in detail, and will also explore the techniques and programming concepts using Apache Storm. Moving on, we’ll familiarize you with “Amazon Kinesis” for real-time data processing on cloud. We will further develop your understanding of real-time analytics through a comprehensive review of Apache Spark along with the high-level architecture and the building blocks of a Spark program. You will learn how to transform your data, get an output from transformations, and persist your results using Spark RDDs, using an interface called Spark SQL to work with Spark. At the end of this book, we will introduce Spark Streaming, the streaming library of Spark, and will walk you through the emerging Lambda Architecture (LA), which provides a hybrid platform for big data processing by combining real-time and precomputed batch data to provide a near real-time view of incoming data.
Table of Contents (17 chapters)
Real-Time Big Data Analytics
About the Authors
About the Reviewer

Optimizing Storm performance

To be able to optimize the performance of Storm, it's very important to understand what are or could be the performance bottlenecks. If we know the pitfalls, only then can we avoid them. Another noteworthy aspect of Storm, like all other Big Data frameworks, is that there is no rule of thumb for performance; every scenario is unique and so the performance optimization plan for every scenario is also unique.

So this section is more about pointers and classic dos and don'ts. What will actually work in your use case to serve as performance booster will have to be uncovered after several rounds of tweaks to the system and after observations.

Fundamentally, Storm is a high-performing, distributed processing system. The moment the work distribution comes into play, it brings with it its own Pandora's box that can be performance glitches such as the interaction between different processes on the same node, or different processes on different nodes that require channels...