nanog mailing list archives

RE: Fire protection in ISPs and collocation facilities


From: Sean Donelan <sean () donelan com>
Date: 5 Nov 2000 23:10:24 -0800


On Sun, 05 November 2000, Roeland Meyer wrote:
Simple rule, make sure that you're upstream and the down-stream end has more
bandwidth than the up-stream end. It's kind of the reverse of the internet.
In the event that the inflow exceeds the outflow, hit the SCRAM switch
before you fry something. This is the main reason I have it (SCRAM), but I'm
damned if I can justify it for that reason. It's a lot easier to talk about
fire. It should be left in the Fire codes.

In 120 years of the Bell System (and post divesture experience) the telco's
have not had a single death caused by a fire in a CO.  The telephone companies
have the concept of service continutity: its better to burn the switch to the
ground before cutting power and service.  This is a very different concept
from NFPA 75 which has an assumption underlying the code, computer hardware
is expensive and must be saved.  In the communications world, the service is
expensive and must be saved; not the hardware.

One interesting tidbit is the NFPA doesn't have a single record instance
of an EPO switch being used by any fire responder.  However there are
hundreds of reports of accidental and erronous EPO activations.

Fuses, circuit breakers or overcurrent devices are satisfactory for cutting
power in every other type of occupancy.  What is so different, and dangerous
about a computer room which requires a manually activated system?  Almost
all the other NEC safety devices are automatic, and activate when required
instead of relying a human.  There are APC UPSes stuffed under desks in
offices all over the country, without EPOs.

If you want an EPO, great; build one.  I'm not proposing banning them.
But don't force me to put one in.  You use the example of the SCRAM button,
however the code specifically exempts systems used for things such as a
nuclear reactor SCRAM system from the EPO requirement.  If I'm building
a reactor, I don't need an EPO; but if I'm building a computer room I
have to install an EPO.

We want to build highly reliable, highly available systems; we need to
look at single point of failures.  EPO systems are designed to be a single
point of failure.  So far I have not seen any data which supports their use
as a life-safety device, and justifies the real damage they do.


What are the four seasons in California? Earthquake, Drought, Fire, and
Flood. You've no doubt heard about the "Oakland Hills Fire" but no one has
ever talked about the mud-slides, in the Winter following the "Oakland Hills
Fire", caused by the ground-cover being burnt off. In fact, no one hears
much about California flooding. That's because it is a natural part of the
ecology here and not as sexy as earthquakes and brush fires.

Every part of the country has its own set of hazards.  Volcanoes in the
northwest, heavy snow in the northeast, hurricanes in the south east,
tornadoes across the mid-west, tsunomi in alaska and hawaii.  Wasn't it
a SPRINT CO which got enveloped by a mud slide a few years ago?  And one
of the few times I lost an AT&T circuit was due to mud slides in California.





Current thread: