[Beowulf] removing tiles around heavy racks?

David Kewley kewley at gps.caltech.edu
Tue Jan 3 16:28:11 EST 2006


On Tuesday 03 January 2006 09:53, Mark Hahn wrote:
> > I've not studied the physics of it, but I can't see where the
> > surrounding tiles would have an effect.  You obviously don't want
> > to
>
> sorry!  I should have explicitly said I'm worried about stability,
> not pure downward load-bearing-ness.  I mean, the floor is level,
> but not microscopically, and if someone leans on a rack, I'd like
> not to have the whole machineroom suddenly shear ;)
>
> > remove the tiles the rack is sitting on.  And those tiles should be
> > supported by a metal grid.  As far as I know, the metal grid and
> > the
>
> presumably the grid (stringers) do most of the lateral stabilization,
> but they're not all that hefty themselves.  obviously with all/most
> tiles in place, the mesh is MUCH stiffer.
>
> but I guess that's the real point: it's a question of how much
> lateral thrust might possibly be encountered, and how well, say, a
> tile-less mesh of stringers+feet would handle it.

Here's how I think about it.  This is after talking to a local scientist 
who's also a consulting seismic engineer and a user of my cluster.

Are the stringers anchored to a load-bearing (concrete) wall?  If so, 
pure compression force into the wall should be OK.  The stringers will 
transmit the force to the wall, and the wall in turn to the ground, 
which is what the feet are anchored to.  So you won't tend to shear the 
floor off the feet.

The first failure mode under such a load would probably be stringer 
buckling, and that'd take a lot of force.  (Assuming your stringers are 
something like mine: basically steel C-channels, about 3/4 inch square 
cross-section, 2 feet long).  Yes, leaving the tiles in will greatly 
reduce the risk of buckling, but for normal forces, buckling is pretty 
unlikely anyway, I'd think.

Tensional force (pulling away from the wall) would test your method of 
attachment to the wall, and possibly the screws holding the stringers 
to the feet as well.  If you don't have the stringers anchored to the 
load-bearing wall, that could be a problem...  We don't (yet).

If your stringers are attached to / hard-up-against a soft wall, you'll 
be OK for any human-scaled forces, but accidents with large equipment 
might cause problems, along with earthquakes of sufficient lateral 
acceleration.

If your stringers are *not* trapped by or attached to walls, then you'll 
be more subject to toppling the raised floor catastrophically, although 
I'd imagine a properly installed floor would be highly resistant to 
human-scale forces.

I think you should be OK removing your tiles, if everyone in the room is 
educated about it and exercises appropriate caution.  But of course 
don't assume I'm right about that.

You know, of course, that a toppling full rack can easily kill someone?  
At least, that's the wisdom I've received, and it totally makes sense 
to me.

Here's how to think about earthquake forces.  What matters is the 
acceleration of the earthquake (three dimensional acceleration), and 
the mass of e.g. your racks.  Basically, the building is accelerated by 
the earthquake, while the racks want to stay put.  So the racks exert a 
force F=ma on the building.  This force can be pretty huge, so you 
really need to think about the buckling limits of your bracing beams, 
and the load limits of your wall attachment anchors.

And you need to think about both the top *and* the bottom of the racks.  
Right now, only the tops of our racks are braced, and in a sufficiently 
large earthquake, the bottoms could pendulum out from under the braced 
tops and the racks crash to the ground.  We have probably 40-50k pounds 
(mass) of joined racks.  In a .5G earthquake, that means 20-25k pounds 
(force) transmitted to the walls.

Yes, I'm a bit nervous about it.  Plus it'd be mighty embarassing for a 
supercomputer owned (in part) by a world-famous seismology department 
to fall over in an earthquake...

David
_______________________________________________
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