A friend of mine from my IBM days (an expert in Data Warehousing, BI, etc.) told me about the Hadoop conference he attended in San Jose few weeks back. When he attended the same conference two years ago in New York, there were hardly 200 attendees whereas this time, the number exceeded 2000 and it was a sold out event. This just proves how fast Hadoop has generated interest. He said that one theme in every presentation was the need for Hadoop skills as almost every presentation had a slide, “we are hiring”.
Hadoop offers a massively scalable data management and analysis environment that can handle many different data types without the complicated transformation and schema changes required to load diverse data into a conventional RDBMS. Remember the days of ETL (Extraction, Transformation, Loading) when data massaging and cleansing preceded the creation of the Data Warehouse for analytics purpose. Given the growth in data volume, velocity and variety, the era of “Big Data” has started and new tools such as Hadoop is the need of the hour for doing search and analytics.
Three vendors are worth mentioning here in the Hadoop solution space.
– Cloudera is the market share leader and it offers the open source Apache Hadoop software (CDH4) in its fourth generation and its proprietary system management software. The new version of CDH offers high availability, improved security and hot failover for the NameNode (metadata server) of the HDFS (file system). This node has been known as single point of failure (not good for enterprise needs).
– Hortonworks, which spun out of Yahoo last year has released its first product Hortonwork Data Platform. It uses Hadoop 1.0 code base (more stable) reassuring the enterprise users. It provides the high availability and failover needs with VMware virtualization and uses open source software for management console and also for ETL (Talend software).
– The third player is MapR which pitches its Hadoop distribution as a high-performance alternative replacing HDFS with a derivative of the Unix-based network file system that is highly scalable and has high availability features. MapR also is part of the Amazon’s Elastic MapReduce service.
Hadoop scales in linear fashion to solve the data-volume challenge and runs on commodity hardware (less expensive). It has challenges in terms of skill shortage and batch-related delays. Many IT shops want to integrate old-school BI systems that are integrated with Hadoop to analyze data inside a cluster or result sets moved out of Hadoop. New Analytics vendors are popping up. Two start-ups are worth mentioning – Datameer and Karmasphere.
Datameer’s analytics platform provides modules for data integration to sources from mainframe to Twitter. It provides a spread-sheet driven data analysis environment meant for business analysts without IT skills. Karmasphere also provides reporting, analysis, and data visualization on Hadoop. It uses a graphical interface and collaborative workflow that works with Hive, the data warehousing component of Hadoop.
Hadoop integration with current BI environment will be a critical need, as years of investment in BI and analytics will not be thrown away to accommodate the new analytic tools.