HPs BDRA - What is different from Traditional Hadoop Architectures?

[#1]

This article will shield light on the HP Big Data Reference Architecture. Hadoop Architectures have been traditionally build to move the math to the data. This has been the basis for Hadoop architectures since the initial implementations back at Yahoo. Let's take a look at traditional verse HP's Big Data Architecture approach.

Current traditional Big Data approach

• Compute and storage are always collocated

• All servers are identical

• Data is partitioned across servers on direct-attached storage (DAS)

tradhadoop.png

New HP Big Data approach

• Separate compute and storage tiers connected by Ethernet networking

• Standard Hadoop installed asymmetrically with storage components on the storage servers and yarn applications on the compute servers

hpbdra.png

With this new approach, HP engineers have challenged the conventional wisdom that compute should always be co-located with data.

HP has been aggressively benchmarking with this new archiecture, and feel they have designed an architecture that will provide maximum elasticity for Big Data Workloads. They are leveraging YARN labels and have not changes a line of Hadoop code with their deployments. Results are significant and have outperformed traditional architecture in a number of tests. This architecture allows for optimized data analytics by running multiple applications while consolidating multiple data stores in a single, high performance system. The HP BDRA is able to be flexible to rapid change, and offers an opportunity to lower TCO.

I would encourage you to read more abou this at HP and Hortonworks web sites.

I hope you find this information useful in your hadoop journey!

Mark Lochbihler

Hortonworks Partner Engineering

@MarkLochbihler