Submitted via IRC for TheMightyBuzzard
MapD Technologies Inc., one of a group of select companies that offer GPU-accelerated databases, today announced the open sourcing of its MapD Core database. The company is contributing the project to the open source community and placing its code on GitHub under an Apache 2 license in order to seed a new generation of data applications.
"MapD pioneered the use of graphics processing units (GPUs) to analyze multi-billion-row datasets in milliseconds, orders-of-magnitude faster than traditional CPU-based systems," the company said in a statement. "By open sourcing the MapD Core database and associated visualization libraries, MapD is making the world's fastest analytics platform available to everyone."
[...] The company also announced a free Community Edition of its software -- provided for non-commercial development and academic use...
Source: https://adtmag.com/articles/2017/05/08/mapd-gpu-database.aspx
(Score: 0) by Anonymous Coward on Monday May 15 2017, @03:57AM (3 children)
Isn't DBMS generally I/O bound? Does leveraging GPU actually help?
(Score: 2) by coolgopher on Monday May 15 2017, @04:24AM (2 children)
It all depends on how you design your schema, what and how you query it, and how large the whole thing is. But yeah, on the "serious" end of the scale, i.e. once you can't keep your indices in RAM, you're looking at an I/O bottleneck typically.
If you only have a few gig of data (or indices) and are running query-heavy things, then maybe there's a place for a GPU-based approach. According to their readme in the github project, mapd-core is a fully in-memory database and there is no rationale listed nor suggested environments it would be ideal for.
(Score: 0) by Anonymous Coward on Monday May 15 2017, @04:37AM (1 child)
Proprietary.
This sounds like a total non-starter open source project.
What is the point of this? We already have non-portable RDBMSes.
We already have open source visualization clients for those other RDBMSes.
We already have years of experience with those others not pulling the rug out from under us, and whether or not they just dumped the code on us and ran away.
So what do these guys actually offer?
(Score: 2) by coolgopher on Monday May 15 2017, @05:03AM
> So what do these guys actually offer?
Buzzwords?
(Score: 2) by kaszz on Monday May 15 2017, @05:09AM
So one will need a system'd CentOS 7 (x86_64) to run the prebuilt package containing dependency packages and "It is preferred, but not necessary, to install CUDA and the NVIDIA drivers using the .deb using the instructions provided by NVIDIA".
You will end up with military industrial complex ruled CentOS that has been system'd thanks to Poettler dependent on the proprietary whims of NVIDIA. In other words enjoy it while you can and don't expect any upgrade path or avoiding planned obsolescence by NVidia or RedHat.
(Score: 2) by Soylentbob on Monday May 15 2017, @07:14AM
... looked fishy for me, given the project is published under Apache2 license. In case you wonder as well:
announced a free Community Edition [..] includes the MapD Core and MapD Immerse visual analytics client. The company also unveiled the MapD Analytics Platform Enterprise Edition, which adds in the MapD Core GPU rendering engine along with distributed scale-out, high availability capabilities -- such as LDAP and ODBC -- that aren't included in the open source version.
Only the core database is open-sourced.