By Peter Jeffcock-Oracle on Mar 05, 2013
I've been to a number of big data trade shows over the last year, and without fail I have the same conversation with many different people. It goes something like this.
We discuss Oracle's Big Data Platform and I mention the Big Data Appliance (BDA). "Oh, yes" they say. "That's a great looking machine, but we can build a Hadoop cluster much cheaper than that, so we're not interested."
The first thing I do is ask them what kind of cluster they are building. They always say something like "I can get 40 $5K servers in a rack for $200K".
"But that's not an equivalent cluster," I will say. The most important number in Hadoop clusters is the amount of storage. When was the last time you heard somebody talk about a 400 core Hadoop cluster? They always say how many terabytes (or even petabytes) their cluster can store. Those smaller servers often only have a few TB of storage, compared with 36TB on each BDA node. So we quickly establish that their equivalent cluster is no such thing. Often it would actually take 2 or 3 such racks to match the capacity of the Big Data Appliance and their "equivalent" system is much more expensive than they thought.
But it's not just about buying servers. When you buy an engineered system you're also getting the rack, the cables, the switches, pre-installed software, tuning, optimization, integrated support and so on. Add those into the picture, and the Big Data Appliance is much lower cost. Take a look at this ESG white paper that goes through all the numbers in detail. Here's the key segment from the executive summary:
"Based on ESG's modeling of a medium-sized Hadoop-oriented big data project, the preconfigured Oracle Big Data Appliance is 39% less costly than a “build” equivalent do-it-yourself infrastructure. And using Oracle Big Data Appliance will cut the project length by about one-third."
If you're building a Hadoop cluster, or looking to expand an existing one, you should keep Oracle Big Data Appliance on your shortlist and give it a closer look.