Jimen Ching wrote:
Its all a matter of risk management.
--jc
And that's the pearl and a half of wisdom thats keeping from jumping up
and yelling "can do". We're looking at several different potential
systems, including the possibility of building our own.
Ex. 1: the ml300 board from Xilinx. It's powered by a ppc405 (I think,
I haven't had any coffee yet, so it could be gibberish) that's enclosed
by a Virtex2 fpga (the whole chip is called the Virtex II Pro). It's
their reference board, and it mostly has everything we'd need on it.
Except that they only have two os's ported to it so far: VxWorks, which
I'm told is not an option, or MontaVista's Linux. I think the board
maybe costs $5k, and with it you get the tool chain for the processor,
but you don't get the liscense or ability to compile and run anything on
the MV Linux stuff. If you want to work with the MV Linux stuff, you
have to spring for the MV Linux dev liscense, which is $25k for up to 3
developers. And that's their cheapest option.
So, questions come up: for $25k + say $10k of engineering time getting
familiar with the board, we'll have a fully functional reference design
and could start development.
vs.
The unknown cost of engineering time it would take to bring up the board
and get everything we need working to the point we could start development.
If I could get it all working in two months, we'd be saving money. But
what if it takes 8. There's a lot of stuff on it. MV had a whole team
of guys working closely with Xilinx to get the board up and running. To
date we've got me and 1.5 EE's.
And maybe that's just too expensive either way. We're not putting out a
product we expect will make us a billion kajillion dollars. There may
very well be reuse on other instruments, but no guarantees of that.
I've done some kernel programming. I've done a lot of "exploratory"
programming on the kernel as well. I think I have a handle on many of
the software mechanisms, and know where to start looking for the
software stuff I haven't directly worked on. But in regards to the
interface between the kernel software and the hardware, that's where I
start getting lost. I've done some programming with interrupts and
that's about it. On other os's & not-quite-os's I've done some light
work with that stuff, but nothing heavy duty.
So I know I can do it, I just don't have a good feel for how long it'll
take. And probably that's too difficult for anybody to really define
with a high degree of accuracy.
Thanks, you guys have given me some ideas.
-Charles
ps. In regards to the MV Linux liscensing, it's actually worse than it
sounds: $25k get's you their development stuff for only one
architecture. I'm not sure how clearly they define architecture, but
I'm sure it's going to suck.