CHAPTER 8In-Memory Computing in Hadoop Stack
By now you are familiar with the Hadoop platform, its broader ecosystem, and some of the computation engines on top of it. You have also learned about the benefits and shortcomings of the traditional MapReduce computational framework. One benefit is linear scalability and the ability to process data in parallel, which comes with the cost of over-reliance on the underlying distributed storage. Each stage of a MapReduce job needs to be written into a filesystem that increases fault tolerance. The process of sending data from the mappers to the reducers, or so-called shuffle stage, can take a heavy toll on the network bandwidth at the time when intermediate data gets copied between the nodes.
This chapter will get into more advanced topics of data processing. In it we explore some of the alternative compute engines and computing technologies, which, unlike traditional systems, open up a great number of beneficial breakthroughs and new ways of leveraging legacy platforms.
From the beginning of Hadoop's creation there have been attempts to make the MapReduce computation engine less complex and more available for non-programmer types. The commonly available system for this is Hive, described in Chapter 4. It adds a SQL engine on top of ...
Get Professional Hadoop now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.
