Overview of HDFS Architecture

Introduction to Hadoop Distributed File System (HDFS)

Hadoop Distributed File System (HDFS) is a robust and scalable distributed file system designed to store and process large amounts of data on commodity hardware. It is one of the key components of the Apache Hadoop ecosystem, playing a crucial role in enabling big data processing.

HDFS Architecture

The architecture of HDFS is specifically designed to provide high performance, fault tolerance, and data reliability, making it suitable for processing big data across large clusters of machines.

Components of HDFS Architecture

HDFS comprises several key components that work together to ensure efficient data storage and processing:

  1. NameNode: The NameNode is the central component of the HDFS architecture. It manages the file system's metadata, including information about files, directories, and their respective locations. The NameNode stores this metadata in memory for faster access and maintains a persistent copy on disk.

  2. DataNode: DataNodes are responsible for storing the actual data in HDFS. They store data blocks on the local file system and communicate with the NameNode to report their status periodically. DataNodes also handle read and write operations requested by clients.

  3. Block: HDFS divides files into fixed-size blocks (typically 128MB or 256MB) for efficient storage and processing. Each block is replicated on multiple DataNodes across the cluster for fault tolerance. The NameNode keeps track of these block locations.

  4. Secondary NameNode: The Secondary NameNode assists the NameNode by performing periodic checkpoints of the file system metadata. It merges the edits log with the file system directory image, creating a new checkpoint. This process helps recover the file system metadata in case of a NameNode failure.

Data Flow in HDFS

HDFS follows a write-once-read-many model, where files are written once and are seldom updated. The data flow in an HDFS architecture can be summarized as follows:

  1. When a client wants to write a file to HDFS, it communicates with the NameNode to get information about where to store the file's blocks.

  2. The client breaks the file into blocks and writes them to multiple DataNodes in the cluster. The client also maintains a pipeline of DataNodes for each block to ensure fault tolerance.

  3. As the blocks are written to the DataNodes, they acknowledge the receipt and report the status to the NameNode. The NameNode updates its metadata accordingly.

  4. When a client wants to read a file, it interacts with the NameNode to retrieve the block locations.

  5. The client then directly reads the blocks from the respective DataNodes, enabling parallel processing and efficient data retrieval.

Fault Tolerance in HDFS

HDFS is designed to be highly fault-tolerant, ensuring data reliability even in the face of failures. It achieves fault tolerance through the following mechanisms:

  1. Data Replication: HDFS replicates each data block on multiple DataNodes (typically three copies) across the cluster. This replication ensures that even if a few nodes fail, the data remains accessible.

  2. Heartbeat Mechanism: DataNodes send regular heartbeat signals to the NameNode to report their status. If the NameNode does not receive a heartbeat within a specific timeframe, it marks the DataNode as failed and initiates the replication process for the lost blocks.

  3. Data Integrity Checks: HDFS uses checksums to verify the integrity of data blocks. Whenever a client reads a block, the DataNode sends the checksum along with the data. The client can verify the checksum to ensure the block's validity.


The architecture of HDFS is specifically designed to handle the unique requirements of storing and processing big data in a distributed environment. By distributing data across multiple machines, replicating blocks, and providing fault tolerance mechanisms, HDFS ensures high availability, performance, and reliability. Understanding the basics of HDFS architecture is crucial for effectively working with Apache Hadoop's data processing capabilities.

noob to master © copyleft