Raised Floors

Wayne S wayne.sudol at hotmail.com
Thu May 23 16:28:49 CDT 2019



Sent from Mail<https://go.microsoft.com/fwlink/?LinkId=550986> for Windows 10

From: Ethan O'Toole via cctalk<mailto:cctalk at classiccmp.org>
Sent: Thursday, May 23, 2019 12:19 PM
To: Patrick Finnegan<mailto:pat at vax11.net>; General Discussion: On-Topic and Off-Topic Posts<mailto:cctalk at classiccmp.org>
Subject: Re: Raised Floors

> Purdue's insurer REQUIRES fire sprinklers in data centers.  And it's not
> atypical from what I've heard.  They are all dry, pre-action systems, which
> eliminate most of the danger of it accidentally dumping water.  It's
> unlikely that with a dry pipe, pre-action system water will be dumped
> somewhere that a person is standing without warning[0]. These days,
> equipment is easy to replace (compared to classic systems) given a good DR
> plan and a good insurance provider.

Most systems do air first (pre action) while triggering the PDU/UPS
systems to shut off first but some X seconds later water is coming out.
Out of the sprinkler head that was broken, not all heads?

Fire Sprinkler heads have bulbs with a liquid (many kinds based upon what the temperature is that they expand) that burst when a specific temp is reached thus allowing the water to flow. They operate independently of each other under the theory that the one that bursts first is closest to the fire and should  be allowed to have the maximum pressure and water flow.

> The NEC also requires[1] an EPO system to shut off all power sources
> (including any UPS).  I'm pretty sure that any fire that required a hose to
> be pulled off of a truck would first result in the power being shut off.

I'm sitting in a fairly large multi-tenant datacenter right now and there
are no EPO buttons. There might be somewhere on a UPS or PDU somewhere,
but nothing I could hit. The amount of damage from doing so would be
horrendus. The employees of the facility might have access to something,
but I would hope it's zoned out.
Some systems have the EPO outside the room in a fairly secure location (IE. locked room) to prevent someone accidentally or maliciously setting it off and causing damage. Sounds wrong, but the idea is that if there is a fire the sensors in the room will activate the EPO.  So no one should be manually turning power off unless they are properly authorized,  as shutting it off will cause false positives in other areas, like activating the fire alarm system and autodialing the fire department and possibly the police too.
That said, I have seen a “Delay” button, which, when pressed, would delay  system activation until a few seconds after  it was released.  This was in a really large datacenter where an alarm would sound when activation was imminent and the system would activate 20 seconds later. Someone working on equipment might not have time to get out of the room before activation. The room was large enough and had a smaller  “forms” room and a IBM tech room with separate doors so people could be in those rooms and not react fast enough to get out in time. The delay was so that someone could give everyone enough time to get out of the room before the system activated. Yes, there was a phone next to the button so that maintenance could be called to reset everything if it was a false alarm.

> HPE server in the datacenter released its magic smoke.  Standard sprinklers
> only affect the area where the fire set them off, and don't have to douse
> your whole multi-1000 sq ft data center.

Right, and smoke sets off the smoke alarm. It takes actual heat to burst
the liquid thing that pops and lets the sprinkler head open. It's not
electronically controlled I would think? Not like the movie "hackers"
would portray?

> 2. In our experience, HPE servers tend to (internally) catch fire and
> release smoke at a rate of around 10-to-1 vs the Dell servers we've had.

Dell BIOS/Lifecycle crud is so slow and horrible :-( . Huge turd. Haven't
seen any HP boxes go up. Had thousands of Supermicro boxes at prior gig
and while the power supplies had engineering flaws never really saw one
put out smoke (sample size possibly 5 digit # of servers.)


--
: Ethan O'Toole




More information about the cctalk mailing list