[Beowulf] Re: removing tiles around heavy racks?

Robert G. Brown rgb at phy.duke.edu
Fri Jan 6 10:45:48 PST 2006


On Fri, 6 Jan 2006, David Mathog wrote:

> My gut feeling is that leaving the racks up on their wheels is more
> dangerous than dropping the feet and coupling them to the raised floor.

Not at all clear to me; my intuition would be nearly the opposite.  Free
rolling racks would in the zero-rolling friction limit experience no
transverse force at all as the floor oscillates back and forth
horizontally underneath them.  For example:

I was recently rear-ended in my giant SUV by a tiny contour.  I had just
enough time when I saw the contour skidding towards my tail to take my
foot OFF the brake -- I had some slack in front of me to spare -- and
was freewheeling but stopped when the collision occurred.  I then braked
down the recoil I picked up -- maybe 4 mph (they hit at maybe 15-20).

My car sustained zero damage.  Literally -- not a scratch.  My running
laptop (sitting on the floor of the car as I was on my way into Duke)
was fine -- I'm typing on it.  My neck is unstrained.  The other car's
damage was much more severe, but nobody got hurt and most of the damage
was cosmetic crumpling as my trailer hitch punched through their front
bumper.

If I had been sitting hard on the brake I would have significantly
increased the impact of the collision, as the energy dissipated
gradually by my post-collision breaking and momentum transferred over a
longer time would have had to have gone into either my bumper or hitch
or theirs.

Now, freewheeling MIGHT let racks roll off the platform altogether as
they vibrate around, unless they are loosely coupled to something that
prevents that... or are maybe put in casters that permit a bit of
freewheeling but constrain it to keep the racks within some tolerance of
their original position, or have sideways shock absorbers that gently
push them back into their original position while partly moderating
their coupling to the transverse forces.

> Inertia is going to provide enough force in even a moderate earthquake
> to get the wheels moving relative to the floor and that's going to put
> a huge load on the top brace since the resulting pendulum motion,
> if perpendicular to the top brace, is going to twist that brace.
> If the racks were standing on their feet, or better yet, bolted
> to the floor, the whole thing would tend to move as a unit so the
> stresses between parts should be less. (Assuming the raised floor is
> strong enough to handle the lateral loads.)

Lots of assumptions there, many of which depend on whether there is a
resonance of any sort between the earthquake and the racks'
pendulum-type oscillation, which are very design dependent.  Moving all
the racks as a unit might tune it INTO resonance where freewheeling does
not -- it all depends on the underlying restoring forces and whether or
not the earthquake has frequency components that match.

Personally I think that free wheeling or not isn't the question, it is
whether or not you have shock absorbers (where free wheeling "can" be a
kind of shock absorption, or at least reduce the force coupling between
the floor and the racks).  The shock absorbers that will do the most
good are the ones that are built into the floor underneath the tiles and
out of sight.  Shock absorbers there can damp the earthquake
oscillations so that they don't REACH the racks -- the entire floor
remains relatively stationary and resonances don't build up as much.
This makes your disks and hardware happier even without worrying about
racks falling over -- a rigid coupling to the ground might drive
frequencies inside the hard disk mounts themselves.

> With the wheels down there's a possible failure mode
> where all racks roll out from under the top brace, which
> twists under the 25k lb load, and then the combined
> mass of the racks pulls down the top brace.  That is, the entire row
> of racks might act like one massive pendulum attached to the top
> brace.  Model it as a mass of N rack weights on a lever arm half
> the distance from the top brace to the floor.  Assume .5g lateral
> acceleration in the worst direction.  That's about 4 ft * .5 * 25000 lb
> worth of torque, at least in the initial phase of the motion just
> as the racks start to move relative to the top brace.  Seems like a
> heck of a lot of torque for the brace to withstand.
>
> As for hopping, you're right that if the units were sitting flat on
> the concrete floor it would take >1g acceleration to make them hop.
> However they are sitting on a more complex structure and that can result
> in accelerations that exceed 1g even if ground motions don't, if the
> resonances in that system match frequencies in the ground motions.

Absolutely.  So damp the resonances.  I was able to google up
earthquake-proofed (within reason) raised floor designs with shocks
pretty easily.  Even doing something homemade -- installing over the
counter shocks between the rigid walls and the floating top layer of the
floor -- would probably help a lot with horizontal motion.  To get
vertical damping, though, you really need special pedestals.

I'm sure that companies have been selling this sort of thing to military
nuclear-blast specifications (not to mention earthquakes) since the
early 60's when SAC lived in holes in mountains.  It can only have
gotten cheaper and better since then.

I'd strongly advise NOT doing this yourself.  Even though I fully
understand the physics involved, I'd rely on an engineer who understands
the PRACTICAL applications of that physics and knows, for example, if I
should be more worried about vertical or horizontal motion in most
earthquakes, what kind of design is stable against various induced
oscillation modes, what kind of loads are assumed by and in tolerance
for the damping capabilities built into any sort of design.  A lot of
this is plain old empirical and derived from tables of measurements, and
you just can't duplicate it by thinking about it as anything less than a
full blown problem in physics with numbers and everything and even THEN
I'd like measurements and engineering better than my own
computations...;-)

With that said, have you seen:

   http://www.asce.org/inside/tclee_elpw+comm.cfm

where they say, e.g.:

"
Guide for the Selection, Installation, and Use of Raised Computer Floors

The committee is developing a seismic guide for raised computer floors.
The guide will include a description of various design features of
raised floor systems, review shake-table test results of raised floors,
review earthquake performance of raised floor systems, and identify
features that appear to contribute to good seismic performance. Guidance
will be provided for selecting raised floors, anchorage and restraint of
equipment on raised floors, and retrofitting of raised floors. 
"

Cisco has a guide to rack assembly here:

   http://www.cisco.com/univercd/cc/td/doc/product/ong/15400/454spint/byatt04.htm#1020585

that addresses raised floor designs and equipment rack coupling in
different earthquake zones.  See also the following article:

   http://www.drj.com/drworld/content/w2_065.htm
   http://www.drj.com/drworld/content/w1_111.htm

which has a paragraph or two on earthquake resistent raised floor
designs and a debriefing on what happened to HP's server rooms during
a serious Bay area earthquake.

There is a book:

   http://www.disastercenter.com/Rothstein/cd126.htm

that is written "just for this".  There are engineering conferences.
This is serious business, with lots of data supporting workable
solutions (and revealing UNworkable solutions).  Not a really great
place to DIY.

(And GIYF:-)

    rgb

>
> Regards,
>
> David Mathog
> mathog at caltech.edu
> Manager, Sequence Analysis Facility, Biology Division, Caltech
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
>

-- 
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





More information about the Beowulf mailing list