[Beowulf] Re: removing tiles around heavy racks?

[Andrew M.A. Cater]

On Sat, Jan 07, 2006 at 08:36:45AM +0000, John Hearns wrote:
> 
> > 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.
> 
> I've seen many machine rooms with painted concrete floors. I think they
> are a good idea in terms of cable management - the ceiling is fitted
> with unistrut, and there are cable baskets suspended from this.
> 
I've also seen a large (new) computer hall where everything was spoken
for, with no possibility of later expansion because all space and every
slot was prebooked/prebuilt. Raised flooring was in place (primarily to 
provide space for a mainframe which needed a dam in case of liquid 
coolant spill) and they were having to rethink because they weren't 
going to fix what they first thought they were going to have and were
needing more forced air cooling :(

Overhead / underneath: _ALWAYS_ provide flexibility :)

> Such a setup doesn't provide forced air cooling via vents at the front
> of the racks. I suppose a system of flexible vents, or maybe even better
> a coupling to the bottom of the rack itself would be a good idea.
> 
Cooling is _always_ a problem - and will get worse with time as
node/blade power consumption goes up relative to a fixed space.
A 42U rack full of 21 2U servers is going to be very different 
from a 42& rack full of 14 3& servers each full of 10 Opteron blades
at a very high power density,

> Also from the point of view of hard working system installers, you just
> roll a rack right in there. No need to get it up a sep or a ramp which
> is common in rooms which were not originally purpose-built to have a
> raised floor.

Have seen somewhere where the roof needed to be removed to get large 
kit in :)

Andy