Hi Ralph,
Thanks for the info! The array sizing could be a problem as the value
I'm thinking about using could potentially be over 100 million. This
value is a unique identifier my system is already generating, I was
considering just using it rather than computing a separate vpid. At
least I have the flexibility to use arbitrary unique values, so I can
work with that.
Maury
On 07/29/11 11:11, Ralph Castain wrote:
There are no hard rules, but some practical ones. In theory, you can use any
value so long as it is unique. In practice, that can cause problems as several
arrays are sized by the vpid value. So a really big vpid causes the array to be
sized to accommodate at least that many values.
In other words, the system was designed as if vpids would be continuous,
starting from zero.
This is something I plan to adjust in the not-too-distant future as sparse
arrays of nodes, procs, and even jobs are becoming more common.
On Jul 29, 2011, at 12:06 PM, Maurice Feskanich wrote:
Hi All,
I have a question about the vpid values. Do they have to start with zero or one, and do they have to have no gaps in the sequence? Basically, I'm wondering if I can use any arbitrary integer, as long as they are unique. If I can, is there a maximum value?
Thanks,
Maury Feskanich
_______________________________________________
devel mailing list
de...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel
_______________________________________________
devel mailing list
de...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel
