Re: [bgl-discuss] Fwd: [bgl-support #302] May I ask you two questions?

["'Kamil Iskra'" <iskra@xxxxxxxxxxx>]

On Wed, Feb 07, 2007 at 16:57:52 -0500, Yongzhi Chen wrote:

> (1) So in you opinion, a 2% fluctuation is normal. I really want to make
> sure it.

The only honest answer I can give is "it depends".  It depends on what you
actually do between your two time samples.

2% flunctuation or more is normal in my experiments, which are file-I/O
heavy.

On the other hand, Blue Gene is renowned for reproducibility of its
results, which stems from its extremely light-weight kernel and the fact
that the whole machine is driven by a single clock source.  So, if your
code is embarrassingly parallel (no communication), I would be surprised by
a 2% fluctuation.  If, however, your code is heavy on MPI communication
between multiple process pairs, I would not be surprised.  Performance of
such communication is hardly ever deterministic, and Blue Gene adds its own
non-determinism by being able to route packets on the torus via different
links to avoid congested areas.

> (2) Can you please tell me the minimum partition that can guarantee me a
> torus machine, 256 or above? Thanks a lot.

Andrew already answered this: 512 nodes.

Kamil

- --------------------------------------------------------------------
To add or remove yourself from this mailing list, use the 'notifyme'
command on any BGL machine. To remove: notifyme -n, to add: notifyme -y.