Big Data Technology – About Hadoop and the Distributed File System


The business world is becoming more and more info-driven than ever before. While leveraging data, the major challenge for any business is to handle the explosion of data sources and real-time streaming of various types of data through these. The traditional database management systems are proven to be failures in handling this changing requirement, and fortunately, there is rapid development in newer technologies centering on big data and redefining the ways to deal with super-massive data volumes.


On thinking of Big Data as the solution, the one most important technology to know about is Hadoop, let's explore.


What is Hadoop?

AdvertisementBuild online presence with trusted marketing software (en)


Hadoop is not a database type, and on the other hand, it can be compared more to a featured software ecosystem which enables massive parallel computing efficiency. Hadoop can be counted as an enabler of a unique set of NoSQL-based databases (distributed in nature like HBase). The role of Hadoop is to spread this data volume across many servers with no reduction in terms of performance and speed.


MapReduce is a key component of the Hadoop ecosystem, which is a different computational model that can take massive data processing tasks and spread it across a big number of servers, known as Hadoop cluster. This technology is a real game-changer when it comes to the enormous processing needs in big data management. A volume data management task which may almost take 20 hours in case of an RDBMS may only take hardly 3 minutes on a distributed Hadoop cluster.


As we may expect Big Data to continue down its growth path in the coming years, it can be assumed that Hadoop will remain as the center of it to allow the enterprises to reach to the fullest data potential. Other parallel advancements in terms of data technology also spark an increasing demand for next-gen data technicians who can handle this powerful and constantly building infrastructure.


HDFS (Hadoop Distributed File System)


When it comes to Hadoop administration, what we need to discuss at the first place is about the Distributed File System, which is the default data storage methodology in Hadoop systems. HDFA uses a kind of ‘NameNode,' ‘DataNode' type architecture to enable a distributed file management system, which ensures both high-performance and quick access to data along scalable Hadoop clusters. HDFS is now counted as the integral part of various Hadoop ecosystems, which provides reliable means to manage big data pools and also support in big data analytics.


Working of HDFS


HDFS enables rapid data transfer between various computing nodes. For this purpose, the Hadoop distributed file system is coupled with the MapReduce, which is another programmable framework to be used in data processing. While HDFS gathers input data, it breaks this information further down to separate simpler blocks and then distributes them in a balanced way to several available nodes in the Hadoop cluster and enables parallel processing.


Above that, HDFS also has a high fault-tolerant capacity. A typical file system in this infrastructure will replicate or copy every data many times and then distribute it to various individual nodes, and ultimately placing one copy of it at the minimum on various server racks. The major advantage of this approach, as pointed out by, is that any random data node which crashes at any point in time can be easily found on another location in the cluster. This enables continuous processing even when the data is getting recovered.


You can also see that HDFS follows master-slave type architecture. In such a model, each of the Hadoop clusters contains a NameNode which manages the file systems, and there are supporting DataNodes which manage data storage on various computing nodes. Other HDFS elements support the applications having larger data sets in a connected way.


HDFS architecture centers on the commanding nodes as NameNodes which hold the metadata as well as secondary DataNodes which store the information in multiple blocks. With Hadoop, the HDFS systems can effectively replicate data at a huge scale and spread across the nodes.




HDFS was introduced at Yahoo, which was a part of the ad serving platform of the company and also meets the requirements of the search engine. No, Facebook, LinkedIn, Twitter, and eBay is some of the bigger corporates in web services offering HDFS to underpin their big data applications.


However, this innovative file system is used far beyond these requirements too. For example, New York Times now use HDFS for their large-scale conversions of the image, the tools like Media6Degrees in machine learning and log processing, LiveBet helping in analyzing odds and log storage, Joost to analyze sessions, and the Fox Audience Network for data mining and log analysis, etc. HDFS also acts as the backbone of various open-source data warehousing technologies, now popularly known as data lakes.


As HDFS is usually deployed in huge-scale enterprise implementations, low-cost support for commodity hardware can be a very useful feature for the users. These systems which are meant to run web searches can range further into thousand of nodes and hundreds of petabytes.


Future of Hadoop


Hadoop has already become a crucial part of the new generation analytical ecosystems. The major achievement of Hadoop now is that it enabled the vision of Data Lake. With this, Hadoop has certainly broadened the analytical ecosystem boundaries. Without such a contribution from it, concepts like Social Media and IoT might not have been possible at this scale. The data which we once rejected as useless has also now become useful. In other terms, we can say Hadoop has changed the way how we look at data.


Considering the future of Hadoop, if you look at the usage trends of Hadoop, we can see that it is not yet used in its fullest potential as an end-to-end platform for analytics. In many of the vectors like Retail, Finance, and Telecom, etc., Hadoop is still used in combination with the time-tested traditional RDBMS solutions, which may not change in soon. Hadoop may need considerable investment and time to cross that bar too and become a full-fledged analytical platform. We need to wait and watch how Hadoop which grow further and cover up all these gaps in future.