[Beowulf] Earthquakes and raised floors...

Robert G. Brown rgb at phy.duke.edu
Fri Jan 6 14:53:34 EST 2006


Just to give one more kick to the dying mule, I was googling on out a
bit farther to see if I could locate any really GOOD online resources
for raised floors and earthquakes when I found the following (APC
sponsored) white paper:

   www.copaco.nl/NR/rdonlyres/56AD946C-FE32-48F2-8D4C-D83E016F089D/0/Raisedfloorsinthedatacenter.pdf

Summary:

  Raised floors are obsolete and should pretty much never again be used.

  Period.

Earthquakes are only ONE reason (although a very good one).  As the
article points out and backs up with numbers, NO data center with a
raised floor can honestly claim 5 nines in reliability, since if a
single major earthquake occurs every 100K years the average damage done
to supposedly earthquake-proof data centers in major seismic events of
the last decade drops the mean expected uptime below five nines.

However, it runs down a whole list of reasons for the use of raised
floors in the first place, one at a time, and indicates that
requirements for data centers have evolved to obsolete them.  For
example, at one time the floor was a room-spanning ground (needed to
prevent ground loops on the network).  Now modern cables prevent ground
loop problems out to a radius of 50-100m, and electrical wiring up to
code to a room panel are adequate to mostly prevent problems on close to
the same scale.  Once upon a time systems needed huge bundles of bulky
cabling; now cabling is thin (and possibly is fiber) and can be done
just as easily and far more accessibly in overhead trays.  Once upon a
time cabling and wiring and so on were "invariant" over the expected
lifetime of the data center, so they could be engineered into the floor
and then not touched for many years.  Now network technologies and
system densities and other requirements vary so rapidly that regular
access to cabling, wiring, cooling occurs, which is actually NOT
terribly easy in a raised floor environment compared to many of the
alternatives.

Earthquake proofed or not, mounting your systems directly onto a solid
concrete floor is pretty much always going to be far riskier than
mounting them on a raised floor that CAN collapse (and WOULD collapse in
any event if the concrete floor it is mounted on collapses).

This white paper, by the way, clearly states (without references, but it
states) that "The full load capability of a raised floor is only
realized when all of the tiles are in place.  The buckling (lateral)
strength of the floor depends on the presence of the tiles.  However,
tiles and even rows of tiles are seen routinely pulled in a data center
when frequently required cabling changes or maintenance is performed.
This situation can give rise to unexpected catastrophic collapse of the
raised floor." which answers David's original question (finally).

I think that many raised floor sellers would argue that adequate angled
crossbracing can compensate for removed tiles and is necessary anyway to
achieve acceptable shear strength, but this is just ONE point listed in
the paper (which is pretty powerful -- it runs down the increased fire
hazard, decreased headroom (which kept us from doing raised floor in the
low-ceiling'd physics department basement), physical security (a caged
on top doesn't prevent somebody from coming in from underneath by
crawling over cables and raising a tile unless your cage extends to the
real ground), the fact that the underfloor space is a dust magnet and
makes the room "dirty" after a while, safety (data operators falling
through where a tile is removed or not properly replaced, associated
liability), and COST -- and additional $20 or more per foot on TOP of
the extra costs doing cabling and cooling in the limited underfloor
space.

It's pretty persuasive.  I think that the safest conclusion is that if
you've GOT a raised floor it may not be worth it to yank it out
(although you SHOULD think about doing so -- some of the negatives are
pretty negative).  If you're doing new engineering, though, you're
probably better off engineering (to) a concrete slab type floor of more
than adequate capacity from the beginning, and put your wiring and
cabling and cooling overhead in patterns that make access and frequent
changes easy and cheap.  The document finishes off with some specific
suggestions for cooling distribution patterns that are likely to be
adequate for this approach -- it concedes that flexibility in
distributing cold air from underneath is the ONE place that raised
floors are at all likely to (still) hold an advantage.

Anyway, read it and see what you think.

    rgb

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu


_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list