>Yesterday we had a powerdip of near half a second. Actually, this was some kind of peak, since three monitors burned away (okay, no flames).
>
>At the server side, we have several Linux, NT4 and one Novell; all on the same boxes.
>The NT servers all kept running (note : the NT clients all rebooted).
>
>What could be an explanation for this ? I don't believe in coindidence.
>
>Peter

I think Hilmar is on to it.

If the voltage on a positive logic gate at dips below the 90% voltage level, or a negative logic gate rises above the 10% voltage level this will trigger a logic change of state in the gate, or in gates connected to it, depending on how long the associated circuitry waits for a stable state.

The powersupply in the NT box is apparently more reactive to changes in the voltage level and/or had a slightly greater reserve capacity to adsorb bigger fluctuations than the other boxes. Since NT OS is the same running as either a server or a workstation (changing only a couple of hex bytes converts a desktop to a server or back again) and your NT workstations went down, the fact that the server stayed up would, I believe, have more to do with the hardware than the software. Components from the same batch used to make a PC are rated to fall withing a specific tollerance, for example say 10%. A resistor, R, may be 10K +- 10%. A capacitor, C, may be 50 uf, +- 10%. Now, for a specific combination of RC values (which is a measure of the time of rise or decay), a power supply may have a slower exponential decay curve when power dips or is removed, allowing the logic to remain above the 90% level for a slightly longer period of time than other boxes manufactured from the same batch of components. Good electronic equipment is made with a 'safety' factor of at least 2X - the components are rated to take twice the range of voltage/current/power levels needed for the design of a particular device. I read somewhere a long time ago that RadioShack equipment was designed with a 1.37X safety factor, which cuts it pretty close. Equipment headed for space (military and NASA) is not only rated at 5X (IIRC) but is also tested for hundreds of hours at varying levels within that range before the components are sent into space. Tom, jump in here.
JLK