Date: Sun, 4 Feb 96 21:41 WET From: uhclem@nemesis.lonestar.org (Frank Durda IV) To: hackers@freebsd.org Subject: Re: Watchdog timer Message-ID: <m0tjHnY-000CFIC@nemesis.lonestar.org>
next in thread | raw e-mail | index | archive | help
[15]As an example for a simple system: in the external model, you would simply [15]provide a chassis spine (those metal things on the ends of cards) with [15]two plugs. Inside, you would connect one cable to the RESET connector on [15]the MLB, and the other to either the SCSI activity light or the IDE [15]activity light connector on the MLB or SCSI/IDE card. [18]Michael Smith <msmith@atrad.adelaide.edu.au> writes: [18]Dang. This is a very neat idea (hits self). Thank you. (Takes an Asparagus bow.) [15]Externally, the cables (shielded) would be run to the monitoring box, where [15]a couple of properly tuned NE555 timers would do an excellent job of [18]Hmm. I'd do this internally, presuming that resetting would be enough. [18]It's not actually as simple as you make it sound though; you need a [18]missing-pulse detector (consider the situation where the disk light stays [18]on) and a timer that runs for longer than the 100-second maximum for the [18]555. Two things: first, you must assume that if the activity light stays on for the maximum amount of time, this is a crash. I have several systems that crash from time to time and they can be spotted visually because the IDE access light is stuck on. No panic, just off in la-la land but the IDE light is on solid. The timer, however implemented, must re-arm on rising or falling edge transitions (doesn't have to be both), not on levels to avoid ignoring crashes while disk access was still in progress and not completed. Note that on the IDE interfaces and Adaptec SCSI interfaces I have used, the activity light is illuminated and then de-illuminated for each access although it may appear to be constantly on. The circuit will be fast enough to spot these transitions. Someone should check NCR and other SCSI brands to see if they have similar behavior. Another way to use the same technique is to access a particular drives' activity light, since most large 3.5" drives provide a header for an external LED, in case the one on the drive will not be visible, ie it is installed in a 5.25" bracket. This would let you see the "blinks" for that one drive. Of course, you would have to make sure the daemon tickled that particular drive, but the rest of the principle is the same. [15]Power would come from a wall adapter. If one timer expires and generates [15]a RESET and things go dead again, a second and much longer timer (five [15]to ten minutes) could cycle a relay and kill power to the computer chassis [15]for ten seconds or so, and then let things reboot. [18]This (the power cycling) would add significantly to the cost of the unit; [18]consider a 15A relay and the extra wiring involved in terms of assembly [18]cost. No doubt, and the pricing I mentioned reflects this. Probably I would simply buy load control modules which several companies make. They are meant for traffic control signal applications. They just look for 5V and activate a load, usually using a 10A solid state relay which will handle the inductive load of a switching power supply just fine. Controlling power is clearly not a feature for everybody, but is something that can be easily added or not added since there are so few components and drill-outs involved in the delta. I would probably use it in important systems that aren't staffed 24 hours a day. Frank Durda IV <uhclem@nemesis.lonestar.org>|"If you say bad things about or uhclem%nemesis@rwsystr.nkn.net | them, they will try to get ^------(this is the fastest route)| you. They come from the or ...letni!rwsys!nemesis!uhclem | Death Star."
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?m0tjHnY-000CFIC>