Patrick McFarland escreveu:
>Any reason why this wasnt posted to the list?
Other than gumstix's list or my e-mail client messing it up, no ...
Thanks for warning, forwarding it.
>On Sunday 26 March 2006 19:29, you wrote:
>>Patrick McFarland escreveu:
>>>On Sunday 26 March 2006 17:55, Alexandre Pereira Nunes wrote:
>>>>What I would love the most is the ability to have two or more gumstixes
>>>>buses connected to each other through proper buffering/latching. It
>>>>would allow me to play a little bit with NUMA clusters.
>>>Not sure if Intel's XScale system would allow it. What'd I'd really like,
>>>of course, is the ability to build complex systems by just plugging more
>>>hardware in. Like, a 3D accelerator should be plugged into the primary
>>>CPU bus (the same one secondary cores and memory controller is plugged
>>Xscale (the core) has everything needed, since all you need is fast
>>transfers and provisions to bus contention. Caching is not optimized for
>>the task, but that's the idea behind numa: not having to make the thing
>>as hardware-stricted as SMP, where that would matters the most. It
>>requires some software provisioning (specially on the kernel) and has
>>some bottlenecks, but it allows many things with hardware not designed
>>to be SMP-aware.
>>But that would be for hobby and academic purposes, since the pxa is not
>>the best hardware to do the task (built-in periferal cpu aren't, in
>>general). On the other hand, as part of the gumwad projects, I happen to
>>have about 6 400mhz around and that would be something to try. The
>>interesting about pxa, in spite of it's limitations, is that it is very
>>fast arm core, floating point excluded.
>>All of that is teory, of course, because I unfortunately (shame on me)
>>couldn't get the gumwad online yet, that would be more useful to people
>>than a working numa cluster nowadays.
>>>Also, it'd be really awesome to be able to actually have secondary cores.
>>>Systems like the GP2X have a ARM920 and ARM940 pairing to do multi-core
>>>work. Being able to arbitrarily add additional ARM940s would be really
>>>awesome. (Again, I don't know if Intel's XScale system allows this.)
>>For a numa system, it could work, provided the bus matched (using
>>asynchronous latches would do most of the trick), and the executable
>>environment were carefully designed, i.e. the user space binaries would
>>have to run neutral arm core (non xscale extensions like coprocessor 0),
>>but that's the case for most binaries today: I run them on qemu-arm on
>>my x86 box without much trouble, personally I think that's very
>>impressive to see since qemu-arm (AFAIK) lacks xscale specifics. Any
>>user-space linux arm code (except for THUMB mode) which does soft-fp
>>seems to work fine with it, whether it was compiled to a xscale, to an
>>arm7 or arm9, or anything between. haven't tried do magic stuff with it
>>though, only elementary networking-based programs and linux system
>>>>Also, your idea of having (flat?) 92-pin cables is very cool, it could
>>>>be very neat to play with experimental child boards having such a
>>>>possibility. The current connector set is appropriated to delicated
>>>>products but very fragile for heavy development environment like most
>>>>hobbists will have (well, I even have a worse one at home, though a
>>>>better one at work).
>>>I assume next generation Gumstix products will have plugs designed to cope
>>>with additional physical stress as not to break easily.
>>I certainly hope so...
>>>No. I think the actual planning of the products is done well. One board
>>>for CPU and memory, one board for network communications and mass
>>>storage, one board for gpio breakout, one board for audio, one board for
>>>robotics; and then a few boards that are a combination of these concepts.
>>I'm not impling a substitition here, I think people should have a bare
>>motherboard to do experiments and have them in a safe, clean fashion,
>>even if you have to plug other sub-boards to do stuff like sound and
>>network on a case-by-case basis, you could have i.e. a bareboard where
>>you plug gumstix and them you plug aside netcfs or netmmcs and audistix
>>or whatever, and the thing just worked. And have a option to buy this as
>>a package, or as i previously proposed, as a motherboard with all on it
>>(identical hardware, no sub-boards). I know this has implications, but
>>this would happen people who lacks sportivity to take a soldering iron
>>to get faster product line, and at a second time, after testing it
>>software, it would assemble the real thing using convetional expansion
>>My idea is not mutual exclusive with gumstix current product line
>>semantics, which I like probably as much as you do.
>>Of course, that's me, and having sub-boards with limited functionality
>>made "not that limited" (i.e. the audiostix II as being proposed) seems
>>to please people the most, but I could not agree with them. I for
>>instance would be happier if the simple boards like waysmall stuart had
>>more pads for gpios and a ferrite bar location for powering from usb,
>>and extensions like these. Specially, again, easier acess to jtag access
>>than test points. I agree that for most people, jtag will never be used,
>>but I'm planning a product line and I'm already investigating how to do
>>board self-test and quality control through jtag, the must would be to
>>have the assembled product do that, but I would like to check the boards
>>first as they arrive, and having to build a professional bed (an amateur
>>one, i.e. one which I would have to check three times if the pins are in
>>position and them hold the whole thing by hand and hold my breath while
>>testing takes place) is just not practical. To me, connectors or test
>>points the size of the 20-pin gpio connector below stuart would do it
>>better (I guess there should be more gpios brought of waysmall stuart,
>>btw, since at home [where I'm a hobbist and lack access to test beds and
>>stuff] i discovered that waysmall stuart is a great piece to plug things
>>>The only additional boards I want to see. One is a prebuilt ready-to-use
>>>board that clones all the functionality of the gp2x, xgp, and psp... high
>>>quality 16-bit stereo DAC, 3D accelerator that can be used with opengl,
>>>wifi, MMC, decently sized high quality LCD or OLED screen, a dpad, 8 or
>>>10 buttons, etc; all you need to do is plug it into a gumstix, and get a
>>>working game platform.
>>I like your requests and if I were working at gumtix (i'm not) I would
>>take them into consideration. I only imagined what you would consider a
>>good source of 3d acceleration, since I lack knowledge of what is
>>perhaps available on this area for embedded systems.
>>>The second board is a wrt-in-a-box setup, which I've mentioned before. 2
>>>ethernet plugs, CF card slot, wifi, usb host ability. Just plug in a
>>>gumstix, and you have a LAN+wifi gateway/firewall router.
>>>I see Gumstix going pretty far because of it's Lego-like architecture.
>>>It's just the lack of the more insane expansion abilities keep it from
>>>being a runaway success.
>>That would be the must. I guess the next gen gumstix(es) would make that
>>very easy. A good thing IMHO would be to have good NICs, which consumes
>>less power (and generates less heat) than the ones used today in netduo
>>for instance, and even better if it could be three (or more) of them on
>>a single piece, with good performance if possible. That's because I tend
>>to assemble transparent firewalls (using briding) on moderate complexity
>>environments and i would need at least three physically isolated zones
>>(intranet, dmz, and outside nets).