Historical data is now an essential tool for businesses as they struggle to meet increasingly stringent regulatory requirements, manage risk and perform predictive analytics that help improve business outcomes. While recent data is readily accessible in operational systems and some summarized historical data available in the data warehouse, the traditional practice of archiving older, detail-level data on tape makes analysis of that data challenging, if not impossible.
Active Archiving Uses Hadoop Instead of Tape
What if the historical data on tape was loaded into a similar low cost, yet accessible, storage option, such as Hadoop? And then data virtualization applied to access and combine this data along with the operational and data warehouse data, in essence intelligently partitioning data access across hot, warm and cold storage options. Would it work?
Yes it would! And in fact does every day at one of our largest global banking customers. Here’s how:
Adding Historical Data Reduces Risk
The bank uses complex analytics to measure risk exposure in their fixed income trading business by industry, region, credit rating and other parameters. To reduce risk, while making more profitable credit and bond derivative trading decisions, the bank wanted to identify risk trends using five years of fixed income market data rather than the one month (400 million records) they currently stored on line. This longer time frame would allow them to better evaluate trends, and use that information to build a solid foundation for smarter, lower-risk trading decisions.
As a first step, the bank installed Hadoop and loaded five years of historical data that had previously been archived using tape. Next they installed Cisco Data Virtualization to integrate the data sets, providing a common SQL access approach that made it easy for the analysts to integrate the data. Third the analysts extended their risk management analytics to cover five years. Up and running in just a few months, the bank was able to use this long term data to better manage fixed income trading risk.
Big Data remains one of the hottest topics in the industry due to the actual dollar value that businesses are deriving from making sense from tons of structured and unstructured data. Virtually every field is leveraging a data-driven strategy as people, process, data and things are increasing being connected (Internet of Everything). New tools and techniques are being developed that can mine vast stores of data to inform decision making in ways that were previously unimagined. The fact that we can derive more knowledge by joining related information and recognizing correlations can inform and enrich numerous aspects of every day life. There’s a good reason why Big Data is so hot!
This year at Hadoop Summit, Cisco invites you to learn how to unlock the value of Big Data. Unprecedented data creation opens the door to responsive applications and emerging analytics techniques and businesses need a better way to analyze data. Cisco will be showcasing Infrastructure Innovations from both Cisco Unified Computing System (UCS) and Cisco Applications Centric Infrastructure (ACI). Cisco’s solution for deploying big data applications can help customers make informed decisions, act quickly, and achieve better business outcomes.
Cisco is partnering with leading software providers to offer a comprehensive infrastructure and management solution, based on Cisco UCS, to support our customers’ big data initiatives. Taking advantage of Cisco UCS’s Fabric based infrastructure, Cisco can apply significant advantage to big data workloads.
Undoubtedly Big Data is becoming an integral part of enterprise IT ecosystem across major industry verticals, and Apache Hadoop is emerging almost synonymous with it as the as the foundation of the next generation data management platform. Sometimes referred to as Data Lake this platform serves as the primary landing zone for data from across a wide variety of data sources. Traditional and several new application software vendors have been building the plumbing -- in software terms data connectors and data movers -- to extract data from it for further processing. New to Apache Hadoop is YARN which is pretty much an operating system for Big Data enabling multiple workloads -- batch, interactive, streaming, and real-time -- all coexisting on a cluster.
The Hortonworks Data Platform combines the most useful and stable versions of Apache Hadoop and its related projects into a single tested and certified package. Cisco has been partnering with HortonWorks to provide an industry leading platform for enterprise Hadoop deployments. The Cisco UCS solution for Hortonworks Data Platform is based on the Cisco UCS Common Platform Architecture Version 2 for Big Data – a popular platform for Data Lakes widely adopted across major industry verticals, featuring single connect, unified management, advanced monitoring capabilities, seamless management integration and data integration (plumbing) capabilities with other enterprise application systems based on Oracle, Microsoft, SAS, SAP and others.
We are excited to see several joint wins with Hortonworks in the service provider, insurance, retail, healthcare and other sectors. The joint solution is available in three reference architectures, Performance-Capacity Balanced, Capacity Optimized and Capacity Optimized with Flash – all support up to 10 racks at 16 servers each without additional switches. Scaling beyond 10 racks (160 servers) can be implemented by interconnecting domains using Cisco Nexus 6000/7000/9000 series switches, scalable to thousands of servers and to hundreds of petabytes storage, and managed from a single pane using the Cisco UCS Central.
New to this partnership is Hortonworks Data Platform 2.1 which includes Apache Hive 13 which significantly faster than previous generation Hive 12. We have jointly conducted extensive performance benchmarking using 20 queries derived from TPC-DS Benchmark – an industry standard benchmark for Decision Support Systems from the Transaction Processing Performance Council (TPC). The tests were conducted on a 16 node Cisco UCS CPA v2 Performance-Capacity Balanced cluster using a 30TB dataset. We have observed about 300% performance acceleration for some queries with Hive 13 compared to Hive 12. See Figure 1.
Additional performance are improvements expected with the GA release. What does this mean? (i) First of all, Hive brings SQL like abilities – SQL being the most common and expressive language for analytics -- to petabyte scale datasets – in an economical manner (ii) Hadoop becomes friendlier for SQL developers and SQL based business analytics platforms (iii) Such performance improvements (from Hive 12 to 13) makes migrations from proprietary systems to Hadoop even more compelling. More coming. Stay tuned !
Figure 1:Hive 13 vs. Hive 12
Disclaimer: The queries listed here is derived from the TPC-DS Benchmark. These results cannot be compared with TPC-DS Benchmark results. For more information visit www.tpc.org.
By now it is clear that big data analytics opens the door to unprecedented analytic opportunities for business innovation, customer retention and profit growth. However, a shortage of data scientists is creating a bottleneck as organizations move from early big data experiments into larger scale adoption. This constraint limits big data analytics and the positive business outcomes that could be achieved.
Click on the photo to hear from Comcast’s Jason Hull, Data Integration Specialist about how his team uses data virtualization to get what they need done, faster
It’s All About the Data
As every data scientist will tell you, the key to analytics is data. The more data the better, including big data as well as the myriad other data sources both in the enterprise and across the cloud. But accessing and massaging this data, in advance of data modeling and statistical analysis, typically consumes 50% or more of any new analytic development effort.
• What would happen if we could simplify the data aspect of the work?
• Would that free up data scientists to spend more time on analysis?
• Would it open the door for non-data scientists to contribute to analytic projects?
SQL is the key. Because of its ease and power, it has been the predominant method for accessing and massaging data for the past 30 years. Nearly all non-data scientists in IT can use SQL to access and massage data, but very few know MapReduce, the traditional language used to access data from Hadoop sources.
How Data Virtualization Helps
“We have a multitude of users…from BI to operational reporting, they are constantly coming to us requesting access to one server or another…we now have that one central place to say ‘you already have access to it’ and they immediately have access rather than having to grant access outside of the tool” -Jason Hull, Comcast
Data virtualization offerings, like Cisco’s, can help organizations bridge this gap and accelerate their big data analytics efforts. Cisco was the first data virtualization vendor to support Hadoop integration with its June 2011 release. This standardized SQL approach augments specialized MapReduce coding of Hadoop queries. By simplifying access to Hadoop data, organizations could for the first time use SQL to include big data sources, as well as enterprise, cloud and other data sources, in their analytics.
In February 2012, Cisco became the first data virtualization vendor to enable MapReduce programs to easily query virtualized data sources, on-demand with high performance. This allowed enterprises to extend MapReduce analyses beyond Hadoop stores to include diverse enterprise data previously integrated by the Cisco Information Server.
In 2013, Cisco maintained its big data integration leadership with updates of its support for Hive access to the leading Hadoop distributions including Apache Hadoop, Cloudera Distribution (CDH) and Hortonworks (HDP). In addition, Cisco now also supports access to Hadoop through HiveServer2 and Cloudera CDH through Impala.