Stop Thinking, Just Do!

Sungsoo Kim's Blog

Apache Spark and Apache Tez

tagsTags

27 April 2014


Article Source

Apache Spark and Apache Tez – Near Real-Time Analysis of Data at Rest

A buzzword for Big Data is “Real-time” and every product is stating they are capable of doing it. But “real-time”, just like “Big Data”, is defined by the user and what their use case and requirements are.

Apache Spark and Apache Tez focus on analysis of data at rest in a near real-time fashion. They still require user intervention to access the data at rest like running MapReduce or a SQL-like query, but have optimized it so it’s not something you need take a nap while you run your job because it’s going to take hours to get your results.

These two projects have done a lot to help you get value out of your data more quickly, but most importantly, they have tackled the two areas that slow down queries the most; spin up time of the processes in the clusters needed to run the job and avoiding unnecessary writes to HDFS.

Here is a break down of these 2 projects.

Apache Spark

  • Creators founded company called “Databricks” and received 14mil from Andreessen Horowitz in it’s first round of funding.
  • Cloudera supported
  • Created to address the performance issues of MR and be a high performance, low latency replacement for it
  • Created before YARN
  • Has been ported to successfully to YARN and running in production
  • Due to Apache Spark being a MR replacement, it improves it’s ability to move outside of the normal MR functions and expand it’s set of possible analyses over HDFS
  • Utilizes a tool called “Shark” which makes Apache Hive Spark compatible to give you the ability to run SQL-like queries

How It Fits: Apache Spark

Apache Tez

  • Hortonworks created and supported
  • Created to be run natively on YARN
  • Created to address the issues of MR and be a compliment to it. Apache Tez Improves the performance of existing MR jobs allowing MR devs to utilize it with little to no learning curve
  • Due to Apache Tez being able to support MapReduce, existing tools like Pig and Hive, that generate MapReduce jobs, will see a performance boost without having to modify their functionality
  • While Apache Tez is made to optimize MR out of the box, it does have the ability to replace MR in the long run as it’s functionality is extended

How It Fits: Apache Tez

TL;DR (Too Long; Didn’t Read)

There are a lot of products, projects and companies that are trying to solve near real-time analysis of data at rest but Apache Spark and Apache Tez seem to be the most promising. While these two projects are tackling the same issue, due to the fact they are going about it differently (Spark replacing MapReduce and Tez complimenting MapReduce) I think there is room for both, at least for the time being.

Want To Know More?


comments powered by Disqus