MapD is a GPU accelerated database with visualization tech built in.
The premise of MapD appealed to me immediately. From the MapD website:
A key advantage of using MapD is that we let you interface with the database with the SQL you know (and our Immerse visualization frontend) while letting us focus on the behind-the-scenes optimizations necessary to ensure that a wide variety of workloads run lightning fast.
Even as your data grows to a size that makes other tools slow to a crawl, MapD allows you to instantly interact with up to billions of records.
However having used MapD a bit now, I have to unfortunately report a few caveats. The first is the SQL engine is immature. I cannot use the “SQL I know” because I know a lot of SQL that MapD cannot run.
MapD can operate on the GPU as promised. But as far as simply writing SQL and expecting it to run lightning fast, well, I have no doubt they are focused on it, but that is not present reality.
So, my first impressions are this company has created an impressive technical feat. I like their style and their direction. They are worth watching as they develop the product.
They clearly cannot be a general purpose database, and they don’t claim to be one. They don’t support such basic features as update statements, and even too many individual insert statements can cause performance problems.
So it means you are getting the data to feed MapD from somewhere else, most likely your current RDBMS.
As an extension to your current RDBMS, there could be a workload at the correct size, where it is large enough to be challenging to process with CPU and system memory, but small enough, of course not to exceed the MapD hard limits, whereby it would be feasible to pass a task or two to MapD for a quicker result. Especially if the task involved a challenging visualization.
That is why I’m going to integrate MapD and Oracle and pass it a visualization task, in a later blog.
Personal aside: my motherboard is fried. I’m not going to have performance metrics until Friday.