Paper showing Linpack scalability of mainstream clusters
Carlos O'Donell Jr.
carlos at megatonmonkey.net
Sat Sep 29 22:02:39 PDT 2001
> > In many (european) countries, the "efficiency factor" -- or "cos(phi)"
> > as we call it -- must be over 0.9 . A transformer, or computer power supply,
> > must be designed to comply with this, at least under normal working
> > conditions. An overloaded transformer won't.
> David's number was also 0.9, so I was misremembering, and it is pretty
> close to one.
Since PC supplies are really nothing more than switching supplies,
they generally have good power factor (the cos(phi) in the equation).
What they also have is good harmonics ;) Your PC will generate higher
harmonic distortions that make all the power supplied to your systems
ugly. Infact most of the load is on the peak of the waveform and tends
to flatten out the voltage supplied.
If you have delicate scientific instrumentaiton, do not run it on the
same circuit as any of your cluster!
Large companies with many computers and odd devices, may need to hire
a power systems engineer to clean up all the _bad_ power devices.
In Canada, anything under 0.9 starts getting you charged extra.
More information about the Beowulf