From owner-freebsd-hackers Sun Feb 4 19:11:37 1996 Return-Path: owner-hackers Received: (from root@localhost) by freefall.freebsd.org (8.7.3/8.7.3) id TAA25346 for hackers-outgoing; Sun, 4 Feb 1996 19:11:37 -0800 (PST) Received: from fw.ast.com (fw.ast.com [165.164.6.25]) by freefall.freebsd.org (8.7.3/8.7.3) with SMTP id TAA25339 for ; Sun, 4 Feb 1996 19:11:30 -0800 (PST) Received: from nemesis by fw.ast.com with uucp (Smail3.1.29.1 #2) id m0tjHIo-000859C; Sun, 4 Feb 96 21:09 CST Received: by nemesis.lonestar.org (Smail3.1.27.1 #20) id m0tjHBc-000CEsC; Sun, 4 Feb 96 21:01 WET Message-Id: Date: Sun, 4 Feb 96 21:01 WET To: hackers@freebsd.org From: uhclem@nemesis.lonestar.org (Frank Durda IV) Sent: Sun Feb 4 1996, 21:01:48 CST Subject: Re: Watchdog timer Sender: owner-hackers@freebsd.org Precedence: bulk [15]By going to PCI, you will ADD for a small-volume board (<1K units) [15]roughly $40 per board JUST for the PCI, both for fabrication, electronics [15]and mechanical packaging. Those PCI chipsets aren't cheap in low [15]volumes, and the packges they come in eliminate do-it-yourself [15]manufacturing. [16]I note that you assume small scale manufacturing in the PCI numbers, but: [15]If you go back to the original task of this "device", which was to somehow [15]monitor the lifesigns of a PC and if there are none for a certain [15]amount of time, RESET the machine, and perhaps if that doesn't work, [15]cycle power on the machine too, a board (even an external box) could be [15]built in small volumes (100 to 500 pieces) to do this job and sold for a [15]reasonable return for around $30US. ($50 for power cycling) [16]You assume large scale manufacturing in the reset board numbers. Uh, less than 1K pieces is "small" and 100 to 500 pieces is also "small". I see no large scale manufacturing here. [16]I really find the idea of using the controller activity light as a [16]timer reset mechanism *very* interesting. Thanks. It is the way I visually check for local systems being down. It seems the IDE light is on solid when several of the machines just decide to lock up. The device would be looking for transitions on these lines, so if it was stuck on or off, it would still trigger a reset. The point of this simple testing mechanism was to provide the necessary functionality at a price that would generate enough volumes to have the board designed and do enough panels to pay for the drill programming and other associated charges in making the actual PCBs. The first five hundred or so are the most expensive. I have done a "100" unit build before and a run that short is really expensive. It hard to recover the board manufacturing costs. [16]I also really, really like the idea of a card edge connector with [16](for instance) a phone jack externally and two cable and two [16]replacement cables for the disk activity lights, and presumably, [16]two connectors for the existing disk activity lights to connect [16]to. I would not use a RJ11 to avoid running into Part 68 issues (and the extremely dreaded German telephone rules which hate RJ11 connectors used for anything other than telephones. I felt that two 1/8" stereo headphone jacks would be best. The main reason is that the RESET cable must be shielded. Too many main logic chip sets (particularly some from Intel such as Neptune and early Triton) are extremely sensitive to noise on the RESET line and I don't want it to trigger just because of what someone wraps the cable around something bad on the inside or the outside of the computer chassis. A second reason for using stereo jacks is that they can be mounted to a metal spine without a supporting circuit board, unlike a RJ11 jack which must be mounted to something. I'd also use only Tip and Ring (not sleeve) so that there would be no potential for shorting RESET or the activity light to the frame at the spine/connector. The Frame/spine would provide the sleeve/shield ground and be unterminated at the external box. I'd probably use a dual-polarity opto for the RESET line, or a low-noise relay, since some PC designs use pull-down and others use pull-up reset controls. We need to keep that circuit isolated and polarity independent. Same with the signal from the drive activity light. It must be isolated since there is no guarantee that we have common ground references between the light and reset signals and our power source. Even with an internal solution, I would still put a stereo jack on the back of the chassis that could trigger an external relay for cutting power to the computer for forced power resets. Again, only tip and ring would be used to maintain isolation. A small battery would be required on internal solutions that cycle power. [16]Actually, you should be able to use a riser socket, like the original [16]8086 "clocks" used and plug into the CPU slot or something else as [16]a molded plastic part with a socket on top and a male counterpart [16]on the bottom (my riser idea). Risers/pass-thrus on BIOS ROM sockets are completely evil for several reasons and I can speak with complete authority on this, having tried to sell products that relied on these little suckers. 1. Some systems do not have the vertical clearance, even if the lowest-profile components are used. 2. Male DIP socket pins are expensive and have to be manually soldered. Thus two-passes of solder work. 3. Not every MLB has the same BIOS package: DIP, PLCC, LCC, etc. 4. Some systems have BIOS ROMs with 16 bit organizations rather then 8. 5. New MLBs even solder the flash part to the board. 6. People get the things plugged-in off position in any of the package types and blow things out. They can't see under the board to get good alignment. 7. Unless you happen to know that the designer or the MLB brought more address lines than needed to the package, you have no additional address space decoding, so you must do something messy such as what SmartWatch does: look for a sequence of accesses on the address lines, and when you get a match, take over all or part of the address space. This adds a lot of complexity to the pass-thru design. If the reason for using one of these is to provide replacement BIOS routines then why not do it by having a plug-in board with BIOS mapped on a 2K boundary somewhere betwen 640K and 1Meg and get it over with? This will be a lot cheaper to make and you can guarantee that it will actually plug into any PC computer you come across. Riser cards will never reach this level of compatibility. [15]buying audience for what is already a low-volume device, since a lot of new [15]machines only come with one or two PCI slots, and it seems we can [15]demonstrate that a lot of FreeBSD users (and potential buyers of this [15]gadget) don't have PCI slots at all or they are full. [16]Actually, most machines are 3 slots, and that's because of line driver [16]limitations in the early chips. Look at any of the "consumer" or "SOHO" grade computers on the market today from any of the top ten makers. You will find 2 PCI the norm and 3 to be almost unheard of except at the very high-end. If you limit yourself to only systems that people put together themselves or the highest of the high-end, the volume for such a product falls too low. [16]There are many boards in Computer Shopper now that are 4 slot boards [16](and have been since October 1995). I agree, but most pre-built machines in the retail channel have two slots or less. [16]I don't run any large service sites, which I see as your primary [16]market. But I do know people who do run large service sites, and [16]they would be interested in the tiny board with the reset button and [16]disk activity light -- it would be damn cheap, and it would be [16]applicable across a wide variety of hardware. The biggest win is that [16]there is no real device driver work, so it's applicable to any OS: [16]BSD, Linux, SCO, UbixWare, Solaris, etc.. Well, that was the idea. (Attention, this publication should be considered to be Prior Art, so no naughty patenting this technique, OK?) The moment you clog a design up with a bunch of extra things that only a few people want, no one is interested in it anymore. You can make the design so that the extras can be optionally stuffed, but that increases the basic design and board drill cost. Each hole and via in the board increases the cost. [16]If you could package it in such a way as to not have it "floating around" [16]in the case, I think you have a winner. Probably a patch of 3M double-stick pad, the thing they used to use to stick AUI converters or AppleTalk interfaces to the back of the computer would be the easiest way to do it for most computers. Cable ties would be needed for use in internal chassis solutions where the design is wire-frame and there are no solid surfaces to stick the thing to. Personally, the better solution is external since out there you can control power and that would be helpful for "extremely stuck" systems, and would let you leverage the ability to control power on more than one computer or external modems, etc. Frank Durda IV |"I'll huff and I'll puff or uhclem%nemesis@rwsystr.nkn.net | and I'll get promoted." ^------(this is the fastest route)| - Old management saying. or ...letni!rwsys!nemesis!uhclem | (C) 1984 FDIV.