Tyler Reed wrote:
> My company is considering a project with other commercial and academic organizations. We work in
> the field of Reconfigurable Computing (RC). RC is basically computing with Field Programmable Gate
> Arrays, which allows for potentially huge performance increases in algorithms that are
> computationally intense. This increase is achieved primarily through paralization of algorithmic
> functions at a very fine-grained level, in hardware. The project we are considering is to 1)
> Identify performance bottlenecks in MySQL, 2) Prioritize the bottleneck list based on projected
> improvement numbers for a parallel hardware version and 3) implement the hardware acceleration
> on an instance of MySQL running on a Cray XD1 supercomputer (which has reconfigurable hardware
> embedded in it).
> Who, within the MySQL community, would be best to discuss this with? Is anyone currently profiling
> the code to identify key bottlenecks? Would there be any way to include a MySQL developer in such a
I've noted your msg to Brian Aker, our director of architecture.
It's a very interesting topic, and it'd be cool to have MySQL involved with this.
I'm just wondering whether an RDBMS would be the optimal place to play with RC.
Here's my thinking:
RDBMS bottlenecks are primarily I/O bound.
After good schema design and decent queries, RAM is the most interesting for caching/buffering, temporary tables, etc - and then of course disk I/O (with seek time being the critical point there, not RPMs or whatever).
In short, I doubt whether optimizing algorithmic functions would gain a significant benefit.
On the other hand, you may just see that as a challenge ;-)
> Hope to hear from you soon,
> Tyler Reed
> Koan Corporation
Arjen Lentz, Exec.Director @ Open Query (http://openquery.com)
Remote expertise & maintenance for MySQL/MariaDB server environments.
Follow us at http://openquery.com/blog/