On Jun 23, 1:04 pm, Mike Treseler <mike_trese...@[EMAIL PROTECTED]
> wrote:
> Norman Bollmann wrote:
> > I've got a software implementation in
> > ANSI-C for a complex database searching. The database is a proprietary
> > format where I am saving data, which has to be given as a result,
depending
> > on the input data. Problem is, the software implementation is far to
slow.
>
> Get a faster server, load linux.
> FPGA's are OK for data filtering or statistics.
> If you need "complex database searching" you
> need a computer.
>
> -- Mike Treseler
The bottleneck for most searches has to do with how many compares can
be done per unit time, which usually boils down to how fast data can
be brought into the search. Unless you have a significantly faster
mechanism to access the data than the CPU, you probably won't be able
to search it any faster. That's assuming you have the undivided
attention of the CPU. If the CPU is busy doing other things while your
FPGA could be doing the searching, that might improve overall
performance.
Block ram usually won't help much, since you can only access one or
two addresses at a time. If you had a multi-word key, then pulling the
data into a multi-way structure (flops or replicated block rams) such
that the different parts of the key could be compared to multiple data
words in parallel, then you'd be getting somewhere.
Andy
|
