Hadoop and Hive/Spark in Data Vault 2.0

In this article, you’ll receive an overview of what Hadoop and Hive is and why they can be used as an alternative to traditional databases.

Hadoop

Hadoop is used to process and analyze large volumes of data efficiently by distributing the workload across a cluster of commodity hardware, enabling parallel processing and providing fault tolerance through its distributed file system and resource management framework.

HDFS – Hadoop Distributed File System

HDFS is a distributed file system that provides reliable and scalable storage for big data. It breaks large files into blocks and distributes them across a cluster of commodity hardware. HDFS ensures data reliability and availability through data replication.

Yet Another Resource Negotiator – YARN

YARN provides a flexible, scalable resource management framework for Hadoop, enabling a variety of applications and workloads to coexist and efficiently utilize the cluster’s resources. It abstracts the underlying infrastructure and allows for the dynamic allocation of resources based on application requirements.

MapReduce – MR

MapReduce is a programming model and processing framework for distributed data processing in Hadoop. It allows for parallel processing of large datasets by dividing the workload into maps, reducing tasks. Map tasks process data in parallel and the output is combined and reduced to produce the final result.

Hadoop Common

Hadoop Common provides libraries, utilities, and infrastructure support for the other components of Hadoop. It includes common utilities, authentication mechanisms, and interfaces that are used by various Hadoop modules.

What is the benefit?

Scalability
Hadoop enables the storage and processing of massive amounts of data by scaling horizontally across a cluster of commodity hardware. It can handle petabytes of data without sacrificing performance.

Distributed Computing
Hadoop distributes data and processing tasks across multiple nodes in a cluster, allowing for parallel processing and faster data analysis. This distributed computing model enables efficient utilization of resources and enables high-performance data processing.

Fault Tolerance
Hadoop provides fault tolerance by replicating data across multiple nodes in the cluster. If a node fails, data can still be accessed from other replicas, ensuring data reliability and availability.

Cost-Effectiveness
Hadoop is designed to run on inexpensive commodity hardware, making it a cost-effective solution for storing and processing large volumes of data. It eliminates the need for expensive specialized hardware.

Flexibility and Extensibility
Hadoop’s modular architecture allows for integration with various tools and frameworks within the Hadoop ecosystem, providing flexibility and extensibility. It supports a wide range of data processing tasks, including batch processing, real-time processing, machine learning, and more.

Data Locality
Hadoop’s distributed file system, HDFS, aims to bring the computation closer to the data. By processing data where it is stored, Hadoop minimizes data movement across the network, reducing latency and improving overall performance.

Ecosystem and Community
Hadoop has a rich ecosystem with a wide range of tools, libraries, and frameworks that extend its functionality for different use cases. It also has a large and active community of users, developers, and contributors, providing support, resources, and continuous improvement.

These benefits make Hadoop a powerful, popular solution for handling big data, enabling organizations to efficiently store, process, and gain insights from vast amounts of structured and unstructured data. The whole ecosystem can also run on-premise, which can make it a good alternative if ‘cloud’ is not an option.

HIVE

Hive is a data warehouse infrastructure built on top of Hadoop that provides a high-level SQL-like query language called HiveQL for querying and analyzing large datasets.

What are the components?

Data Storage
Hive leverages Hadoop Distributed File System (HDFS) as its underlying storage system. It stores data in HDFS in a distributed and fault-tolerant manner, allowing for scalable, reliable data storage.

Schema Definition
Hive allows users to define a schema for their data using a language called Hive Data Definition Language, like DDL. This allows users to define tables, partitions, columns, data types, and other metadata associated with the data.

Query Optimization
Hive optimizes queries by performing query planning and optimization techniques. It aims to generate efficient query execution plans to minimize data movement, optimize resource utilization, and improve query performance.

Hive Metastore
Hive maintains a metadata repository called the Hive Metastore. It stores information about the tables, partitions, schemas, and other metadata associated with the data stored in HDFS. The metastore allows for efficient metadata management and retrieval during query processing.

Extensibility
Hive offers extensibility through User-Defined Functions (UDFs), User-Defined Aggregations (UDAs), and User-Defined Table Functions (UDTFs). These allow users to define custom logic and operations in programming languages like Java, Python, or other supported languages.

Integration with other tools
Hive integrates with various other tools and frameworks in the Hadoop ecosystem. For example, it can work alongside Apache Spark, Apache Pig, Apache HBase, and other components to provide a complete data processing and analytics solution.

Partitioning and Bucketing
Hive supports data partitioning and bucketing, allowing users to organize and store data in a structured manner. Partitioning involves dividing data into logical partitions based on specific criteria, while bucketing involves dividing data into equally sized buckets based on hash values.

SerDe
Hive uses a serialization/deserialization framework called SerDe (Serializer/Deserializer) to read and write data in different formats, such as CSV, JSON, Avro, and more. Users can specify the appropriate SerDe for their data format to ensure proper data processing.

Overall, Hive simplifies data querying and analysis on Hadoop by providing a familiar SQL-like interface. It abstracts the complexity of writing low-level MapReduce or Tez jobs and provides a declarative and user-friendly approach to interact with large-scale data stored in Hadoop.

Conclusion

Hadoop is a robust and feature-rich environment that can be challenging to manage. However, its numerous advantages make it a compelling choice, depending on the user’s needs and the available in-house expertise. If you’re interested in learning more about it, watch the following recording.

Data Vault 2.0 on Microsoft Azure

In this newsletter, you’re going to get an overview of what real-time processing is and what possibilities it can provide your Data Vault 2.0 implementation.

What to expect

You will learn that real-time processing gives you the ability to create value out of data quicker, have the most up-to-date data in your reporting tools and allow more accurate decisions regarding data.
With that, your company will be able to adapt to changes in the market quicker by seeing developments right away with the most recent data.

Additionally, you can save costs by moving away from batch loading because the peak of computing power normally required for that gets reduced and is more evenly distributed throughout the day. That is especially the case when using cloud environments, because then it’s possible to replace promised environments and contribute the needed computing power perfectly.

The traditional way – batch-loading

Batch loading is a traditional method used to load data into a data warehouse system in large batches, mostly overnight. The data from data sources is delivered up to a certain time in the night to be transformed and loaded into the core data warehouse layer.

This method leads to a peak of data processing overnight, and organizations have to adjust their infrastructure needs to be able to deal with the expected maximum peak of required computing power.

The new way – real-time data

Real-time data is processed and made available immediately as it is generated, instead of being loaded in batches overnight. When using real-time approaches, the loading window is extended to 24 hours. So the overnight peak and its disadvantages are gone.
When using real-time data, it’s always modeled as a non-historized link or as a satellite.

Possible use cases for real-time data are vital monitoring in the healthcare industry, inventory tracking, user behavior on social media or production line monitoring.

Different types of real-time data

There are different types of real-time data based on how frequently the data is loaded and the degree of urgency or immediacy of the data.

Near real-time data refers to data that is loaded in mini-batches at least every fifteen minutes, with the data stored in a cache until it is loaded into the data analytics platform.
Actual real-time data, also called message streaming, involves loading every single message directly into the data analytics platform without any cache.
This type of real-time data is useful when it is important to have data available as soon as it is generated for dashboards or further analytics.

The acceptable processing delay for real-time data is typically defined by the consequences of missing a deadline. Additionally, there are three types of real-time systems: hard real-time, soft real-time, and firm real-time.

Implementing real-time processing

So, how do you implement real-time data processing into your data warehouse solution? There are many architectures for that, but we will focus on the Lambda and Data Vault 2.0 architecture.

The lambda architecture separates data processing into a speed layer and a batch layer. The speed layer processes real-time messages with a focus on speed and throughput, while the batch layer provides accuracy and completeness by processing high volumes of data in regular batches. The serving layer integrates data from both layers for presentation purposes.

At first, the Data Vault 2.0 architecture seems to be similar to the lambda architecture, but it treats some aspects differently. The lambda architecture has issues from a Data Vault 2.0 perspective, such as implementing a single layer in each data flow and lacking a defined layer for capturing raw, unmodified data for auditing purposes.

The Data Vault 2.0 architecture adds a real-time part called “message streaming” to the existing batch-driven architecture, with multiple layers implemented for capturing and processing real-time data, integrating it with the batch-driven flow at multiple points. Messages are pushed downstream from the publisher to the subscriber, loaded into the Raw Data Vault and forked off into the data lake. But the main process is the push inside the message streaming area. The architecture is able to integrate data from batch feeds or to stream the real-time data directly into the dashboard.

Using Microsoft Azure for real-time processing

Microsoft Azure is a cloud computing platform and set of services offered by Microsoft. It provides a variety of services, including virtual machines, databases, analytics, storage, and networking. These services can be used to create web and mobile applications, run large-scale data processing tasks, store and manage data, host websites and much more.

The illustration describes a typical real-time architecture used by Scalefree consultants, which follows the conceptual Data Vault 2.0 architecture.

Data sources deliver data either in batches or real-time. This is loaded into the Azure Data Lake or accepted by the Event Hub beforehand. The Raw Data Vault Loader separates business keys, relationships and descriptive data using Stream Analytics and forwards the message to the Business Vault processor. The Business Vault processor applies transformation and other business rules to produce the target message structure for consumption by the (dashboarding) application. The results can be loaded into physical tables in the Business Vault on Synapse or be delivered in real-time without further materialization in the database. The target message is generated and sent to the real-time information mart layer implemented by a streaming dataset, which is consumed by PowerBI. The cache of the dashboard service will expire quickly, but the Synapse database has all data available for other uses, including strategic, long-term reporting.

Conclusion

In conclusion, real-time data processing offers numerous benefits over traditional batch loading methods, including the ability to create value out of data quicker, have the most up-to-date information in reporting tools, and make more accurate decisions. By adapting to changes in the market quicker, companies can stay ahead of the competition. Moving away from batch loading can also save costs by reducing the peak of computing power required.

As mentioned before, the last illustration shows an architecture that the Scalefree Consultants implemented to make use of real-time data.

Read more on our recently released Microsoft Blog Article.

How is your current experience with real-time data processing?
Are you thinking about kick-starting your Data Vault by also using real-time data?
Or are you already using it and thinking about improving it further?

Let us know your thoughts in the comment section!