Cloudera’s Doug Cutting delivered a presentation at Hadoop World that outlined key forces driving the data world forward which shed some important insights on where enterprise technology is going. This presentation is summarized and embedded below:
Doug starts with a review of his models for predicting the future and this is important, since every model has strengths and weaknesses. His model for predicting the future of data is to take a linear approach. We know what has happened in the past and can know what is happening now so can make linear predictions on the near term future.
For example, with hardware the past has always been that hardware gets cheaper and transistor counts go up so this cheaper hardware gets more powerful. This power and drop in price applies not just to processors but storage. The prediction this leads us to is a simple one, we will be able to store and process more data in the future and will do so more economically. Why do we care? Turns out that in parallel with this growth in capabilities in hardware is the adoption of technology. We have seen technology invade every element of industry and our personal lives and all this generates data. They leave tracks of what we are doing, what our institutions are succeeding and failing at and give us an image of ourselves and our businesses and if we look at the growth in industries they are greatest when they are automated.
Regarding software, in the past decade we have seen remarkable transition towards open source software as platform technologies. Linux and Android are obvious examples of this. The benefits and drivers for individuals and business are becoming very clear. Consider, for example, the fact that business are now built around technology. It is bad to have your business overly dependent on others in ways that increase risk and this translates to strong pushes for open source technologies. You don’t want your business to be hostage to someone else’s business. Open source avoids this lock-in and does so economically. You pay for the value you receive. So what prediction can we make from this? In the future, platforms for data must be open source. It is a basic requirement.
A key prediction can be made about open source in general and Hadoop specifically. A decade ago Doug Cutting began work on the project that became Hadoop. The community came together and made it reliable. The community enhanced its security. High availability was introduced. Improvements continued and continued. So we can generalize and predict from this. The Hadoop platform, already very capable, will continue to improve.
What has been happening around Hadoop? We have seen innovation around this platform. It started as a background component and with frameworks around it became easier to use, then new capabilities like HBase, Impala and Search. It is now very capable. The prediction to make: more and more types of workloads will be supported.
What does this mean for us all? The fact is that Hadoop now dominates in the Big Data space. It is no longer the place you come to to process data, it is the locus of your enterprise data. It lets data be shared and hardware resources be shared. People are taking workloads out of silos and bringing them into Hadoop. We are seeing this used as an enterprise data hub.
On the topic of transactions on data: Transactions have long been thought to be out of scope for Hadoop. Doug believes this is clearly an important class of workload that is not served by the current Hadoop platform. A year ago Google published another paper describing their internal system they built for internal use that does this, demonstrating this is possible. In the past when we have seen this is possible, in two years it happens. The prediction: it is inevitable. It will come to this platform.
In Conclusion: Hadoop is playing its role as the center of an enterprise data hub. We are in the middle of a revolution in data. Revolutions can be scary times. But Hadoop provides a clear path that will endure, supporting a wide variety of workloads. Have no fear. Be comfortable adopting Hadoop.
