Seeking Watchdog Hardware?

hpavc asks: "I have played with watchdog daemons and hacks on and off for a while. I am wondering if there are any reccomendations out there for which real cards watchdog cards are of the best value. Or as some suggest is it all just a scam. From what I read the mechanisms of the cards differ greatly, both in their monitoring and their response. From simply watching an area of DMA to an embedded device with a os that monitors a large amount of things on the computer hardware and the environment. (tia)"

Two hardware methods

[FueledByRamen]

I'm assuming you'd be using this with Linux, correct? If you are, this would be _easy_ (although it would work with any other OS you knew how to write drivers for). Two ways:

1. Your machine must have an ISA slot (or you must know how the hell PCI works). Make a card with an onboard micro (something simple - 68HC11, 8052) connected to an unused IRQ and a chip-select system for an address in the middle of ISA adapter card space. Every 1 second, have the micro trigger an interrupt. The OS should see this if it's still alive and kicking in some fashion. If it doesn't see some pattern of 8 writes to that address space within a short period of time, have it short the hardware reset pins together (plug an unused port on the micro into the reset header on the mobo).

2. Same concept, different implementation. Hook a micro (BASIC stamp would be perfect for this, cheap and easy to program) up to the serial port. If it sees the repeating character stream sent out by the OS, all is well. If it stops for more than 1 sec or so, hardware-reset time!

Both of these would be relatively easy to implement for someone with hardware knowledge, and fairly inexpensive (ISA board would be maybe $20 in parts + whatever an ISA prototyping card costs, serial port version $45 for Basic Stamp kit + MAX232ACPE RS232 level-shifter chip and associated capacitors).

Re:Two hardware methods

[unitron]

"...connected to an unused IRQ..."

Why wasn't this modded +5 Funny?

PCI-WDT500

[hectorh]

I have used the PCI-WDT500 before and it worked just perfectly.

My application was a mpeg stream server using icecast.

I modified the icecast code so that it reset the watchdog timer after some important sections of code. If for any reason the icecast server would not be able to perform those tasks (ie, the mpeg stream was not being transmitted), the watchdog timer would expire and the machine would reboot.

Needles to say, this method would cover all types of problems from bugs in icecast, to kernel panics, hardware failures and network problems.

The PCI-WDT500 is a PCI bus version, so it will work with all newer systems. All you need to install it is to hook it up to the hardware reset line.

Watchdog and timer cards for linux

[bradams]

This link has info on linux watch dogs.
watchdog and timer cards for linux [topology.org]

If all you need is lockup detection

[Yarn]

From /usr/src/linux/Documentation/nmi_watchdog.txt:

On Intel and similar ix86 type hardware there is a feature that enables us to generate 'watchdog NMI interrupts'. (NMI: Non Maskable Interrupt which get executed even if the system is otherwise locked up hard). This can be used to debug hard kernel lockups.

IBM's Automatic Server Restart

[Khazunga]

If you still didn't buy your servers, IBM's xSeries servers come with a watchdog integrated in the motherboard. They call it Automatic Server Restart, and works like any other watchdog. Set it up, call it every x seconds. Fail for n times, and the machine reboots on its own.

If the xSeries has it, I'd imagine the pSeries does too -- but, alas, I have none of those :-P

Re:IBM's Automatic Server Restart

[cpthowdy]

If anyone cares, Compaq, er, HP has ASR also in the Proliant servers.

Watchdog PCI card

[jafuser]

What's wrong with something like this [berkprod.com]?

http://www.berkprod.com/pci_pc_watchdog.htm

Supports Linux and Windows...

Chipset timeout.

[toft]

intel i8xx chipsets (or only some?) have an integraded watchdog timer. Look at the i810_tco driver in linux. It actually works on my ThinkPad to..

Sun boxen

[Gothmolly]

Have this. It's called LOM (Lights Out Management). Full serial manageability, and a hardware reset gizmo on the system board. Run "lom -d 40000,10000" and if the box hangs for 40 seconds, it gets rebooted.