r.watershed

[de584438 at longs.lance.colostate.edu]

last week I wrote with a problem using r.watershed for basin
deliniation.  My elev map is large (rougly 1700 x 750 cells).  I am
using a min outside basin of 150 sq. km.  The problem is that I do not
have enough mem to run up front.  As a result, r.watershed allocates
disk space and swaps to this space within the grass dir. structure.   
The code has been running for 4-5 days ( forgot when I started it), and
is nowhere near to being finished.

Jan Hartman responded with a suggestion of redefining SEG_INDEX ( in
ramseg.h ) as a function.  I did this to no avail.

The system I am on has a large area set aside for disk swapping

shavano.lance: pstat -s

81912k swap configured
19732k reserved virtual address space
        15944k used (4928k text, 96k smem)
        65968k free, 3980k wasted, 0k missing
avail: 2054*32k 240*1k

However, r.watershed apparently does not recognize this.  In fact,
because it is running in 'slow' mode it is only utilizing 1.0 M of memory

USER       PID %CPU %MEM   SZ  RSS TT STAT   TIME COMMAND
de584438  5773  7.8  5.2 1000  600 p0 D N  1473:23 /USH0/GRAD/CE/de584438/grass4

As you can see, this job has been running for a long time, with most of
it spent with disk access.

Can anyone suggest a way for r.watershed to ID the system swap space? 
This in itself will allow the job to run much much faster.

I would also appreciate any input some of the CERL programmers might have.

thanks a lot


Dan Epstein