Tyler Reed wrote:
> Thanks for the response Arjen.
> You are correct in your assertions. But if you think about how a database engine works, there
> exist a number of potential optimizations points. First off, new FPGA's have broken the 10 million
> gate mark and are increasing very quickly from the logic density perspective. Alongside this, most
> platform FPGA's have a large amount of extremely fast, dual-port RAM embedded in the chip. If an
> index were placed in the FPGA (assuming it would fit), a complete index search would take about 1
> or 2 clock cycles. Newer FPGA's are clocking in around 500MHz, so I leave the math to you as
> homework :) Also, there are valid reasons to look at RC as an option to offload and accelerate some
> of the traditional I/O bottlenecks found in traditional computing platforms. Imagine a soft
> processor with direct and simultaneous access to multiple SCSI or SATA disk drives or arrays.
> Combined with algorithmic parallelism possibilities, we believe that at least doing the
> initial performance profiling and acceleration projections merits a closer look.
> What do you think?
Like you said, worth a look. It's definitely interesting stuff.
I suggest you subscribe to our Internals mailing list: http://lists.mysql.com/internals
Then you can run things by some of our developers, ask questions, etc.
Arjen Lentz, Exec.Director @ Open Query (http://openquery.com)
Remote expertise & maintenance for MySQL/MariaDB server environments.
Follow us at http://openquery.com/blog/