[Gate-users] Problem with benchSPECT on a cluster

[Simon Stute]

Hey,

I just think that when you want to split a simulation that is made of
multiple runs (i.e. a timeSlice smaller than timeStop-timeStart), the number
of splits should be set in order to get a virtual time (total time per
split) that is a multiple of the timeSlice.
Let's take an example to be more clear :

timeSlice = 2
timeStart = 0
timeStop = 10

--> you thus get 5 runs

In that way, you have to choose a number X of splits so that [timeSlice * X
/ (timeStop-timeStart)] is an integer.
X=10 --> virtual time = 1 --> [2 * 10 / 10] = 2 --> you get 2 splits for a
run (GOOD)
X=16 --> virtual time = 0,625 --> [2 * 16 / (10)] = 3.22 --> you get 3.2
splits for a run (PROBABLY A PROBLEM)

But it is just a suggestion !
Cheers,
Simon

On Tue, Apr 19, 2011 at 1:25 PM, Tomi Äijö <tomi.aijo at gmail.com> wrote:

> Hi!
>
> I'm using a large-slace cluster (up to 1500 nodes) for Gate simulations. I
> have been testing the setup with bundled benchmarks: The benchmarkPET is
> running just fine, but I have problems with the spect benchmark.
>
> The error message is:
>
> --
> Sorry, but the study duration (x s ) does seem to be a multiple of
> the time-slice (y s). You must change these parameters then restart
> the simulation.
> --
>
> If I have understood correctly, virtual time slices should fix this
> problem?
> The split file indicates that virtual time slices are being used. I'm using
> gatev5_p01.
>
> Am I doing something wrong? Should I modify the benchmark to be able
> run with more than (acquisition time / time slice) nodes?
>
> BR,
> -Tomi Äijö
> _______________________________________________
> Gate-users mailing list
> Gate-users at lists.opengatecollaboration.org
> http://lists.opengatecollaboration.org/mailman/listinfo/gate-users
>