What is Hadoop HDFS?

Have a Database Problem? Speak with an Expert for Free
Get Started >>


When you’re working with very large sets of data, storage can be a real issue. While a relational database can be an adequate solution for more modest data sets, scalability and performance become serious problems when big data is involved. Fortunately, Hadoop comes with a storage solution that can tame even the largest of data sets. The Hadoop Distributed File System (HDFS) was designed with big data in mind– it can store massive amounts of data reliably and offers streaming data access for applications that run on the system. In this article, we’ll learn more about Hadoop HDFS and explore the inner workings of this distributed file system.

How Does HDFS Work?

It’s important to be familiar with a few key terms and concepts in order to understand how HDFS works “under the hood”:

  1. Blocks: A block is a fixed amount of data that the system can read or write. In HDFS, the default size of a block is 128MB, though this value can be configured. Data is split up into blocks in HDFS; the blocks are then distributed throughout the various nodes of the cluster. This distribution is what makes efficient parallel processing feasible.

  2. NameNode: HDFS has a master-slave architecture, where the NameNode plays the role of the master. The NameNode stores the metadata in HDFS, but doesn’t store the actual data itself. It also holds information about blocks and their location for any file found in HDFS, which means it knows how to construct any given file from its respective blocks.

  3. DataNode: In the master-slave design of HDFS, the DataNode acts as the slave. DataNodes hold all the actual data stored in HDFS, and they’re in constant communication with the NameNodes. If a DataNode in a cluster is down, there’s no impact on availability of the data it holds. The NameNode will manage the replication of the blocks needed to account for the data held by the unavailable DataNode.

Once data is stored in HDFS, it can be processed using the popular MapReduce programming model, which is another essential component of the Hadoop ecosystem. This programming model works by first mapping data into key-value pairs, and then reducing it based on user-defined specifications. With this two-step process, even massive data sets can be crunched and analyzed quickly and efficiently.

Advantages of Using Hadoop and HDFS

There are multiple reasons why organizations choose to use Hadoop and HDFS to handle very large data sets. Let’s look at some advantages of this distributed file system:

  • Flexibility: Hadoop can accept data from a wide variety of sources, and the data doesn’t have to be structured.
  • Cost: Hadoop relies on “commodity hardware” for storage, which is a term that refers to affordable devices that are easy to acquire. This means that the cost of adding a new node to a cluster is low, allowing you to scale out without breaking your budget.
  • Performance: The distributed architecture of HDFS allows you to crunch massive amounts of data at top speed.
  • Fault Tolerance: HDFS is highly fault tolerant due to its use of replication. Replicas of user data are created in different machines across a HDFS cluster. If one node in a cluster fails, data that resided on that machine can still be accessed from other machines in the cluster that had a replica of that data.
  • Scalability: Hadoop is designed with a horizontal model of scalability. This means that you can scale “out” by adding more machines to a cluster instead of having to scale “up” with a costly equipment upgrade.


When you need to store and manage big data, a traditional relational database isn’t going to cut it. Hadoop HDFS is a storage solution for big data that offers all the flexibility, scalability and performance your organization requires. While this article merely provides an overview of HDFS and its benefits, it can serve as a good starting point for a deeper dive into research if needed.

Pilot the ObjectRocket Platform Free!

Try Fully-Managed CockroachDB, Elasticsearch, MongoDB, PostgreSQL (Beta) or Redis.

Get Started

Keep in the know!

Subscribe to our emails and we’ll let you know what’s going on at ObjectRocket. We hate spam and make it easy to unsubscribe.