Metis yields 3 partitions when nparts = 8

I have a small test graph, 16 nodes, 32 edges. I ask for 8 partitions, but Metis yields only 3. What's happening? When will I get fewer than the requested partitions? (Input file and gpmetis output after the break)

Here is the input file:


% Metis graph file:
16 32
% edges for each vertex (starting with vertex 1)
3 4 8 12 16
3 4
1 2 4 5 10 14 16
1 2 3 8 16
3 6 7 9 13
5 10 11 13
5 15
1 4 10 12 15
5 10 12
3 6 8 9
6 12 13
1 8 9 11 13 15
5 6 11 12
3 15
7 8 12 14
1 3 4
% last vertex: 16 (should equal )
% last adjacency: 64 (should be 2 * )


Here is the gpmetis command and output (it's really v5.0.1; METIS_VER_ SUBMINOR was unchanged in the bug fix):


$ gpmetis -ptype=kway -minconn test/exp.metis 8
******************************************************************************
METIS 5.0 Copyright 1998-11, Regents of the University of Minnesota
(HEAD: , Built on: Sep 30 2011, 12:39:22)
size of idx_t: 32bits, real_t: 32bits, idx_t *: 64bits

Graph Information -----------------------------------------------------------
Name: test/exp.metis, #Vertices: 16, #Edges: 32, #Parts: 8

Options ---------------------------------------------------------------------
ptype=kway, objtype=cut, ctype=shem, rtype=greedy, iptype=metisrb
dbglvl=0, ufactor=1.030, minconn=YES, contig=NO, nooutput=NO
seed=-1, niter=10, ncuts=1

Direct k-way Partitioning ---------------------------------------------------
- Edgecut: 12, communication volume: 17.

- Balance:
constraint #0: 3.000 out of 0.500

- Most overweight partition:
pid: 7, actual: 6, desired: 2, ratio: 3.00.

- Subdomain connectivity: max: 2, min: 0, avg: 0.75

- There are 0 non-contiguous partitions.
Total components after removing the cut edges: 3,
max components: 1 for pid: 2.

Timing Information ----------------------------------------------------------
I/O: 0.000 sec
Partitioning: 0.002 sec (METIS time)
Reporting: 0.000 sec

Memory Information ----------------------------------------------------------
Max memory used: 0.063 MB
******************************************************************************