Tuesday, February 10, 2015

Ways to import unstructured data into Hadoop

There are multiple ways to import unstructured data into Hadoop, depending on your use cases . 
  1. Using HDFS shell commands such as put or copyFromLocal to move flat files into HDFS. For details, please see 
    File System Shell Guide
  2. Using WebHDFS REST API for application integration. 
  3. Using Apache Flume. It is a distributed, reliable, and available system for efficiently collecting, aggregating, and moving large amounts of data from many different sources to a centralized data store, such as HDFS. Even though historically lots of use cases of Flume are involved with log data collection/aggregation, Flume can be used together with Kafka and turn itself into a real-time event processing pipeline.
  4. Using Storm,  a general-purpose, event-processing system. Within a topology composed of bolts and spouts, it can be used to ingest the event-based unstructured data into Hadoop
  5. Spark's streaming component offers another alternative to ingest real-time unstructured data into the HDFS. Its processing model is quite different from Storm though.  While Storm processes incoming event one at a time, Spark streaming actually batches up events that arrive within a short time window before processing them. It is called mini-batch. Spark streaming of course runs on top of Spark Core computing engine, which is claimed to be 100x faster than MapReduce in memory and 10x faster on disk. :-)

No comments: