We have a problem with DMP and a HP RP4440 connected via 2 fibre switches
to a HP XP128 disk array.

vxdisk list shows both paths to the disk as enabled.
vxdmpadm listctlr shows the paths as being enabled.

We then create some i/o to the disk using dd or cat.
sar -d at this stage shows that the I/O is being shared fairly equally between
the two paths to the disk (c4 and c6).
At this stage we needed to test what hapens in the case of a failure of one
of the channels so we disable one of the ports on one of the switches or
pull out a fibre. This seems to cause immense problems for dmp.
At this stage the remaining I/O channel floods to 100% with extremely long
response times and eventually the disk group fails with failing disks.
vxdisk list shows both paths as still enabled, which would suggest
that dmp has not reacted to the path being taken offline.
If we dd to one of the actual dmp paths, e.g /dev/vx/rdmp/c4t0d1 and disable
the port on the switch, then dmp immediatley reacts by disabling the path
as expected.
Any level higher than the dmp devices though would appear not to respond
(e.g. /dev/vx/rdsk/dg01/fsysname)

Any ideas what is going on please ??