It’s my pleasure to be spending much of this week with hundreds of data-heads, learning about all things data at Dataversity’s Enterprise Data World (EDW) in San Diego. As my focus is on architecture and Cloud Computing, the part of EDW that floats my boat is the Big Data story, especially when the Cloud is involved.
It’s no surprise, then, that Qubole caught my attention. Qubole is a Big-Data-as-a-Service (BDaaS) service provider. I know, I know, yet another *aaS, right? However, in Qubole’s case, it’s not Cloudwashing. They truly have a BDaaS story.
Qubole enables the collection, refinement, and consumption of Big Data sets, offering the power of Hadoop and related Big Data analytics tools running in the Amazon Cloud. But that basic data in -> crunch -> results out equation doesn’t illustrate what’s special about Qubole: Qubole is a service for data analysts and business people, not for developers. It goes beyond what Amazon offers to provide a business-focused BDaaS capability.
Contrast Qubole with Amazon’s alternative, Amazon Elastic MapReduce (EMR). Like Qubole, EMR offers Hadoop as a service. But to use EMR, you need to work directly with Hadoop, which means you need solid Java skills. Remember, Hadoop is a Java framework for writing analytics algorithms, more so than an analytics application itself. With Qubole, however, there’s no need to monkey with Java. The Qubole interface abstracts the underlying Hadoop engine.
Another interesting twist to Qubole is that it works with objects stored in the customer’s S3 object store (that’s Amazon’s Simple Storage Service). Point Qubole to your data and let it go. As a result, Qubole doesn’t have to provide its own data security, as it’s up to you the customer to properly configure S3’s built in security capabilities.
One limitation of Qubole is that because it works directly with an object store rather than a relational database, it doesn’t work well with normalized relational data. You can move relational data to S3 table by table, but you lose relational integrity. That being said, Hadoop itself isn’t designed for relational data, either. Rather, it’s intended to work with a mix of different data types including unstructured data. As a result, so is Qubole.