We’ve all read the articles where someone with a massive data set unleashes the power of Big Data and discovers an awesome and magical insight about their customers that changes the course of history as we know it. We like to call them Magical Big Data Unicorns—so frequently discussed and pursued, but so rarely seen in the wild! We’re not saying they don’t exist, but we would like to suggest a better approach to getting solid ROI from these new-fangled Big Data tools.
For example, when many data gurus think of Hadoop, they think of analyzing huge, unstructured datasets that will uncover important corporate insights. It feels like everyone is marketing the power of Hadoop to find that priceless gold nugget buried within petabytes of social media, blogs, emails, reviews, clicks and chats. Because those stories have great anecdotal power behind them, they can often light the imaginations of business users and help vendors sell new products. Just like a Magical Unicorn.
What is not as frequently advertised and discussed is the ability for Hadoop to help organizations save really big IT dollars in far less dramatic ways. These scenarios don’t make for cool infographics or fascinating Nate Silver data epiphanies, but they are absolutely meaningful techniques for significantly reducing IT costs.
This article examines several scenarios for using Hadoop to save on IT budgets and improve your Service Level Agreements (SLAs). While there are other possibilities of cost savings, this article covers the following scenarios where Hadoop-based cost reductions could be implemented and measured.
Hadoop will not replace an organization’s enterprise data warehouse. However, more and more enterprise data warehouse architects now believe that Hadoop could assume pieces of the data warehouse workload—and with that “joint architecture” come significant cost savings.
As data volumes grow, and data types increase and become more diverse, SLAs for various projects and phases are put into jeopardy. IT staff are constantly squeezing everywhere they can for more time and cycles to get everything done in increasingly tight 24 hour windows. In fact, huge data users like Yahoo, LinkedIn, Facebook, Twitter and Google, became the key drivers for development of the open source Hadoop community in the first place. They simply needed a wider-scale, faster platform. Early adopters at Yahoo tell of transformations that used to take eight hours shrink to 15 minutes due to Hadoop performance capabilities.
Hadoop has the ability to allow IT staff to offload highly compute-intensive manipulation and transformation tasks at a much lower cost than typical data warehousing platforms. Data transformation is an excellent candidate to use the power of Hadoop’s processing to offload data these data manipulation workloads. This frees the data warehouse to more effectively execute queries, perform analytics tasks and generate timely reports.
When a data warehouse is paired with Hadoop, the organization can now avoid the trade-off between valuable transform compute cycles and the value returned from an unencumbered data warehouse.
A key benefit to Hadoop is that companies can take advantage of the decline in storage costs across the board. Hadoop can be a very cost-effective alternative to keep older data out of the data warehouse—but still keep it online. These “active archives” could be a benefit to many organizations that might want or need the ability to query multi-year datasets for patterns and outcomes.
Commercial BI tools are now available to use SQL with Hadoop directly—and these tools will continue to evolve and improve. This will accelerate the utilization and interest in these inexpensive active archives for data mining and other data science activities.
Data scientists and analysts were big users of small data samples to test things like predictive analytics. It was an ideal place to explore data. As systems and needs changed, the size of the sandboxes got bigger so that predictions and other results improved over time.
Now, with the advent of inexpensive, large data stores, sandboxes can be used for real-world testing, analysis and simulation. Using the features of Hadoop, predictive analytics can be run against enormous quantities of data. If your organization has been using expensive platforms for these sandbox systems, huge cost savings can be attained by moving them to Hadoop and its less expensive offline storage.
Data scientists are completely engaged in developing new scenarios that let them now analyze unlimited dataset sizes for increased accuracy in their models.
Hadoop has the ability to significantly lower costs of data management. It will not replace your data warehousing and BI infrastructure. It can help reduce contention for operational resources and allow cost-effective access to active archive data for longer-term historical analysis. One more benefit is the reduced need to continually add costly new technology to the data warehouse simply to meet service level agreements.
Start small when reviewing where it is believed Hadoop can play a role in reducing costs. Find time and resource pain points and see if a solution can be built from that perspective.
Click here: Avoid the Magic Unicorn Syndrome to download a formatted PDF copy of the article.