This week we checked in with a number of thought leaders in big data computing to evaluate current trends and extract predictions for the coming years. We were able to round up a few notable experts who are driving big data innovations at Yahoo!, Microsoft, IBM, Facebook’s Hadoop engineering group, and Revolution Analytics—all of which are frontrunners in at least one leg of the race to capture, analyze and maintain data.
While their research and technical interests may vary, there is a general consensus that there are some major developmental shifts that will change the way big data is perceived, managed, and of course, harnessed to power analytics-driven missions.
Although it’s difficult to present an overarching view of all trends in this fluid space, we were able to pinpoint some notable movements, including a broader reach (and subsequent boost in innovation) for big data and its use; enhancements to data as a service models following increased adoption; and more application-specific trends bound to shape the big data landscape in coming years. As you might imagine, we also considered these movements in the context of opportunities (and challenges) presented by cloud computing.
Widespread Adoption via Increased Innovation
One of the most echoed statements about current trends in the big data sphere (from storage to analytics and beyond) is that as data grows, the tools needed to negotiate the massive swell of information will evolve to meet demand.
According to Todd Papaioannou, Vice President, Cloud Architecture at Yahoo, Inc., and former VP of architecture and emerging technologies at Teradata, the big trend for big data this year will be widespread adoption of via increased innovation, particularly in the enterprise—a setting that has been reticent thus far. He notes that while this push for adoption was not the case for traditional business one or two years ago it is being enabled by the “expanding Hadoop ecosystem of vendors who are starting to appear, offering commercial tools and support, which is something that these traditional enterprise customers require.”
In addition to the richer ecosystem that is developing around big data, Papaioannou believes that much of the innovation in the Hadoop and big data ecosystem will be centered around enabling much more powerful analytical applications.
Of the future of big data, he says that sometime down the road, “people will be amazed at the data they used to throw away because the insights and value they gain from that data will drive key decisions in their businesses.” On that note, there will be more emphasis on this valuable information and its organization. Papaioannou states that companies will have a big data strategy that will be a complimentary piece of their overall general data and BI strategy. In his view there will be no more “islands of data” but at the core there will be a big data platform, which will be at the center of a seamless data environment.
In terms of cloud’s role in the coming wave of big data innovation and adoption, he suggests that cloud will drive down even further the cost associated with storing and processing all of this data, offering companies a much wider menu of options from which to choose.
CTO of Revolution Analytics, David Champagne also weighed in on the role of cloud computing in the push to optimize big data processes, nothing that there are some major hurdles when it comes to large datasets and the issue of data locality. His view is that there need to be effective ways to either collect the data in the cloud, push data up to the cloud, or ensure low latency connections between cloud resources and on-premise data stores.
Like the others, Champagne sees an ever-mounting push toward advanced analytics and optimization as demand for ways to manage large datasets grows. Given the data deluge, Champagne suggests that in 5 years there will be more organizations with practical experience in managing and processing huge data sets, and this will drive optimization. “Working on data as it streams into data stores and finding different ways to aggregate results on data partitions, will be requirements to handle this greater load.”As he states:
“Whether it is Hadoop or Microsoft’s Dryad, we’ve seen the benefits of distributed file systems, distributed computing frameworks, and SQL-like capabilities in those frameworks. The ability to execute statistical operations on and visualize huge data sets is ultimately where the end users of such systems want to go. Being able to run regressions or cross tabulations using huge data sets will offer insights that cannot be achieved by just using small samples. The challenges with something like R are both integrating it into a distributing computing framework, and adapting algorithms to work with independent blocks of data.”
Enhanced Data as a Service: The New “Big Data Marketplace”
While there are likely going to be rapid developments in terms of frameworks and methodologies to handle massive datasets, with the evolution of more sophisticated tools will come an increased demand for even more data—a demand that Roger Barga, Architect with Microsoft’s eXtreme Computing Group feels will be met by a new kind of marketplace.
Barga is a believer in the power of analytics and large data to drive business growth. He admitted that he sees big data computing as one of the most significant innovations in the last ten years—a heady distinction, given the multitude of advances that have emerged.
Barga suggests that there has been a change in the way people think about vast amounts of data; it has gone from a challenging problem to a rich source for any number of business objectives. He notes that Microsoft is beginning to see a “new class of decision makers who are very comfortable with a variety of diverse data sources and an equally diverse variety of analytical tools which they use to poke and prod data sets to unlock new insights.“
In Barga’s view, big data will continue to get bigger as more ways to uncover its value emerge. His prediction is that, due to the insatiable demand for data since there are so many new tools to analyze and handle it, in the coming year we will see the rise of data marketplaces. These will be hubs where decision makers can find the data they need to make decisions, to extract signals and gain valuable insights to help them run their business.
As Barga describes, these “data marketplaces will go beyond simple “data as a service” offerings available today and provide both trusted public domain and premium commercial data in an integrated consumption experience with easy discovery, exploration and purchase. The result will be much more than just datasets but allow decision makers to mash rich datasets with their high-value enterprise data to be more competitive.”
In terms of the future of big data, Roger Barga puts forth a number of lofty goals for the uses and developments in big data computing. He states:
“In the coming years we are sure to see innovative applications in fields from the internet to biotechnology and predictive analytics. Machine learning and analytics over extremely large data is still in its early stages of development and is an active area of research today. The automated or semiautomated analysis of large volumes of data lies at the heart of big-data computing for almost all application domains.
The cloud, big data, and automated analysis, present very interesting opportunities. One of the greatest opportunities is that people can leverage the a la carte economics of elastic computing to carry out machine learning and data analytics that were prohibitively expensive due to the requirements of building and maintaining their own hardware infrastructure. In the future cloud computing may serve as a bridge between big data and predictive analytics models built using cycle-hungry machine learning and statistical based algorithms.”
With a new marketplace to push even more data into the datacenters of even more companies, in-house resources will be (if they aren’t already) stretched to the max. Looking to the cloud for storage is one thing, but as the tools develop, cloud computing could play the bold role that Barga predicts in the nearer (versus more distant) future.
Context Accumulation and Real-Time Analytics
While some have already addressed some of the more framework-related aspects of evolution in the big data space, on the application and use end, this all seems a little abstract. To put the evolution of big data trends in context, IBM Distinguished Engineer and Chief Scientist behind IBM’s Entity Analytics, Jeff Jonas revealed some specific developments, including context accumulation.
Jonas explains this further:
“There is a big difference between a big pile of puzzle pieces versus a puzzle in some state of assembly. Context accumulation is the process whereby each new observation introduced to the enterprise (a new puzzle piece) is evaluated against the work in progress to determine if it has a place, or an affiliation. What has been missing to date, for the most part, is organizations have been attempting to look at (e.g. use algorithms against) individual observations before first seeing how the observation relates to the historical data already gathered. That is kind of like yelling “fire” when one sees a puzzle piece with flames on it — before first taking the piece to the puzzle where one quickly discovers that this puzzle piece is a fire … in a fireplace.”
To Jeff Jonas, context accumulation over big data will also introduce very exciting new properties including “lower false positives, lower false negatives, and the more data one harnesses, the faster one will be able to compute! This exciting reality works in much the same way the last few pieces of the puzzle are as fast and easy as the first few, despite the fact one has more data in front of them than ever before. This will prove to be a game changer for the big data community. “
On that note, there is an interesting exploration of those ideas at Puzzling: How Observations Are Accumulated Into Context
Jonas predicts that over the next five years, the big data community will begin shifting its attention more and more towards real-time, streaming, analytics over big data — not just periodic end-of-day batch jobs, courtesy of Hadoop.
Batch processes provide users insight periodically, e.g., once a day or once a week. However, it will only take so long before they stand up and say, “Hey, why did I have to wait until the end of day? I can no longer use some of this insight, as the customer left the web page or left my store. Can’t I get these high quality outputs in real-time?”
Notably, there will still be a need for periodic introspection on identifying patterns and these patterns are more likely to be found using deep reflection. So the future, I say, will be a mix of real-time and batch big data analytics. Whether these future systems run in public or private clouds, it is quite likely either way it will be in some form of a cloud computing infrastructure … and when I say cloud what do I mean? I mean elastic in response to changes in demand and architecture over grid computing for endless horizontal scaling.”
Big Data Clouds on the Horizon
Aside from IBM’s Jonas, others are sensing how the merger between framework advancements and practical uses is playing out—and interestingly, how the cloud enhances this marriage.
Dhruba Borthakur, a lead Hadoop engineer at Facebook, shared his views on the trends occurring in big data management based, in part, on his experiences with the social networking megasite. Borthakur feels that clouds represent a major part of new developments for large data, noting that as of now, the most common element for big data lies in batch processing systems that crunch big data sets (with the focus being more on system throughput rather than latency). He notes, however, that:
“Going forward, I visualize that more latency-sensive realtime applications will move to big-data infrastructure this coming year. For example, Facebook has started to store all Facebook User’s email messages in a Hadoop based cloud, and latency of requests is a critical factor for this infrastructure. In the coming year, I expect many other companies to start adopting big-data clouds for their real-time applications as well.
According to Borthakur, the next few years will bring about a number of changes for application developers in particular. In his view:
In the current state of affairs, the closest to a traditional database in the big-data world are some key-value stores, e.g. HBase, Cassandra, etc. These existing data stores differ from traditional databases, they do not offer ACID properties. This makes the life on an application developer more difficult. The next five years will see a proliferation of cloud-based applications: to satisfy the demands of these applications, we will start seeing newer big-data technologies that are closer to a real distributed database.
Borthakur predicts that these new databases will scale to multiple petabytes and will provide some flavors of transactional property that will make it easier to write distributed cloud-based applications.
Again, like the others who took part in this survey of opinions, he sees the power of big data to reshape the business and research landscape but notices that there are still some missing pieces that need to fall into place before the ecosystem can become completely self-sustaining.
To conclude, there is no doubt that cloud computing will play an important role in handling big data demands, especially as the predictions about rapid innovations create demand for new ways to store, process and analyze information.