Equally as important as determining Hadoop’s necessity in your business environment is choosing the right Hadoop distribution. Ultimately, you will find that your decision depends on a host of criteria, though performance and scalability are two major attributes that you should examine closely. Let’s take a look at some specific Hadoop performance and scalability requirements, as well as a few key architectural requirements.
Performance
One of the main reasons for moving away from a traditional database solution for managing data is to increase raw performance and gain the ability to scale. It may come as a surprise to you to know that not all Hadoop distributions are created equal in this regard.
In a previous article, How 250 Milliseconds in Added Latency Can Ruin Online Sales This Holiday Season, we take a look at how slower performance (high latency) can directly impact your bottom line. Slow website performance can lead to a decrease in online sales conversions of up to 7 percent, which translates to millions of dollars lost for high volume online retailers.
As you can see in the graph below that compares the MapR M7 Edition with another Hadoop distribution, the difference in latency, and thus performance, between distributions is staggering.
The need for high performance increases even further when you consider the real-time applications of Hadoop, such as that of financial security systems.
Thanks to technologies like Hadoop, financial criminals are finding it increasingly difficult to steal digital assets. Financial services firms like Zions Bank are now able to stop fraudulent financial threats before any real impact is felt by banking customers. Dependability and high performance are essential features for analyzing and reacting to real-time data in order to prevent destructive fraudulent activity.
Scalability
Another primary benefit of Hadoop is its scalability. Rather than capping your data throughput with the capacity of a single enterprise server, Hadoop allows for the distributed processing of large data sets across clusters of computers, thereby removing the data ceiling by taking advantage of a “divide and conquer” method among multiple pieces of commodity hardware.
While this architecture was the beginning of the data scalability revolution, it is by no means the end. Within the Hadoop platform there are three further considerations regarding scalability:
File Bottleneck
The default architecture of Hadoop utilizes a single NameNode as a master over the remaining data nodes. With a single NameNode, all data is forced into a bottleneck. This limits the Hadoop cluster to 50-200 million files.
The implementation of a single NameNode also requires the use of commercial-grade NAS, not budget-friendly commodity hardware.
A better alternative to the single NameNode architecture is one that uses a distributed metadata structure. A visualized comparison of the two architectures is provided below:
Photo credit: Architectural Overview of MapR’s Apache Hadoop Distribution by M.C. Srivas via SlideShare; Slide 58
As you can see, the distributed metadata architecture uses 100% commodity hardware. In addition to the savings in cost, it boasts an equally pleasing 10-20 times increase in performance and avoids the file bottleneck with a file limit of up to 1T, greater than 5000 times the capacity of the single NameNode architecture.
Node Expansion
Smaller users of Hadoop will have smaller data storage and processing requirements and therefore can afford to run on fewer nodes. Larger implementations can find themselves using upwards of thousands of nodes.
This is where the scalability of Hadoop really shines. Going from an entry-level big data implementation to thousands of nodes within a cluster is an easy expansion. Adding commodity hardware as needed minimizes the cost involved in your data processing expenses and allows your investment to grow with your needs rather than ahead of them.
Node Capacity
In addition to the quantity of nodes, Hadoop users should also examine the processing and storage capacity of each when physical storage limitations are a concern. If it is, you can reduce the overall quantity of nodes, while also maintaining data storage requirements, by using nodes with higher disk densities.
Architectural Foundations
Performance and scalability within your Hadoop implementation can be further enhanced by using a distribution that keeps several architectural foundations in mind.
Minimizing Software Layers
Performance within your Hadoop system can be easily obstructed when too many software layers have to be navigated.
Working within a Single Platform for All of Your Big Data Applications
Some Hadoop distributions may require you to create multiple instances. An optimal implementation will allow all workloads to be processed within a single environment. This reduces data duplication which consequently improves both scalability and performance.
Utilizing Public Cloud Platforms for Elasticity and Scalability
A good distribution will give you the flexibility to use Hadoop within your own firewall as well as on reliable cloud environments such as Amazon Web Services and Google Compute Engine.
In the end, selecting the right Hadoop distribution should be less about conforming your business to your selection and more about your selection fitting within your current and future needs. Analyzing the performance and scalability qualifications of each distribution, as well as considering the architectural foundations, is fundamental to a successful evaluation and implementation of Hadoop within your organization.