---
title: "GFS vs HDFS architecture"
subject: "Information System"
chapter: 8
key: "semester_8/information_system:gfs-vs-hdfs"
source: https://www.bctengineer.com/q/semester_8/information_system/gfs-vs-hdfs
site: BCT Engineering
---

# GFS vs HDFS architecture

Information System · Chapter 8: Scalable and Emerging Information System techniques

- Asked in: 2073 Magh, 4 marks
- Video: [Distributed File Systems Explained: GFS vs HDFS vs Ceph | Complete Guide for Beginners](https://www.youtube.com/watch?v=OjvQEod63gg) — CodeLucky
- Answer checked by Claude Opus 5.5 (high effort) on 2026-10-14
- Page: https://www.bctengineer.com/q/semester_8/information_system/gfs-vs-hdfs
- Whole chapter: https://www.bctengineer.com/md/semester_8/information_system/chapter-8

## Question

Compare GFS Vs HDFS and their corresponding architecture.

## Short answer

GFS is Google's proprietary, HDFS is open-source. Both have master (GFS: Master, HDFS: NameNode) and chunkservers/DataNodes. Chunk size: GFS 64MB, HDFS 128MB. Both replicate data thrice. Architecture: client contacts master for metadata, then reads/writes directly to chunkservers/DataNodes.

## Answer

**GFS (Google File System)** is Google's proprietary distributed file system. **HDFS (Hadoop Distributed File System)** is the Apache open-source file system modelled on GFS.

| Basis | GFS | HDFS |
|---|---|---|
| **Owner** | Google, proprietary | Apache, open source |
| **Master node** | Master | NameNode |
| **Slave nodes** | Chunk servers | DataNodes |
| **Unit of data** | Chunk, 64 MB | Block, 128 MB (64 MB in old versions) |
| **Replication** | 3 by default | 3 by default |
| **Writes** | Multiple writers, record append supported | Write-once-read-many, single writer; append supported later |
| **Metadata log** | Operation log and checkpoints | EditLog and FsImage |
| **Heartbeat** | Master and chunk servers exchange HeartBeat | NameNode and DataNodes exchange heartbeats and block reports |
| **Language** | C++ | Java |
| **Processing partner** | Google MapReduce | Hadoop MapReduce |

#### Architecture
```text
 GFS:                      HDFS:
 Client                    Client
  |  \                     |  \
  |  Master (metadata)      |  NameNode (metadata)
  |                         |
 Chunk servers             DataNodes
 [chunks x3]               [blocks x3]
```
In both, the client asks the master/NameNode for locations, then reads and writes data directly with the chunk servers/DataNodes, so the master is not a bottleneck for data transfer.

## Check yourself

The answer's key points as multiple-choice questions, with the correct option and why.

1. Which mapping of GFS to HDFS terms is correct?
   A. Master -> DataNode, Chunk server -> NameNode
   B. Master -> NameNode, Chunk server -> DataNode, Chunk -> Block
   C. Chunk -> NameNode, Master -> Block
   D. Master -> EditLog, Chunk server -> FsImage
   Answer: B. GFS master = HDFS NameNode, chunk servers = DataNodes, chunks = blocks. (Key point: Terminology mapping)

2. Which write model describes HDFS?
   A. Write-once-read-many with a single writer
   B. Many concurrent writers with record append as the main feature
   C. Random in-place updates by any client
   D. Read-only forever
   Answer: A. HDFS is write-once-read-many with a single writer (append added later); GFS supports multiple writers and record append. (Key point: Write model)

3. What are the typical data unit sizes in GFS and HDFS?
   A. GFS 128 MB, HDFS 64 MB always
   B. GFS chunk 64 MB, HDFS block 128 MB (64 MB in old versions)
   C. Both 1 GB
   D. Both 4 KB
   Answer: B. GFS uses 64 MB chunks; HDFS uses 128 MB blocks, 64 MB in older versions. (Key point: Unit size)

4. Why is the master/NameNode not a bottleneck for data transfer?
   A. Clients get locations from it, then read and write directly with chunk servers/DataNodes
   B. All data flows through it but at high speed
   C. It stores no metadata
   D. It is replicated on every client
   Answer: A. Only metadata requests go to the master; data moves directly between client and storage nodes. (Key point: Read/write path)

## Related in Information System

- [MapReduce with GFS and HDFS](https://www.bctengineer.com/question.md?key=semester_8%2Finformation_system%3Amapreduce-gfs-hdfs) — Information System, chapter 8

## Related in other subjects

- [Compare RISC and CISC architecture](https://www.bctengineer.com/question.md?key=semester_4%2Fmicroprocessor%3Arisc-vs-cisc) — Microprocessor, chapter 6, asked 4×
- [GFS vs other file systems](https://www.bctengineer.com/question.md?key=semester_8%2Fbig_data_technologies%3Agfs-differ-other-fs) — Big Data Technologies, chapter 2
- [GFS master role in read/write](https://www.bctengineer.com/question.md?key=semester_8%2Fbig_data_technologies%3Agfs-master-read-write-flow) — Big Data Technologies, chapter 2
- [Stateful vs stateless; NFS architecture](https://www.bctengineer.com/question.md?key=semester_7%2Fdistributed_system%3Anfs-stateful-stateless) — Distributed System, chapter 2