#148 mosmon reports dead for cluster head -- no migration
I have a small ten node cluster that consists of PIII
PCs using Fedora Core 1 with a master "head" using the
same OS but newer 1800+ dual Athlons.
The head has two network cards one primary for the
internet at large and another for the interal 10 nodes
at 192.168.1.
I am using the same release of 2.4.22-openmosix3smp.
The head has an IP of 192.168.1.1, the nodes,
192.168.1.101 through 110.
They can all communicate, SSH, NFS to the head and vice
versa with no issues.
The nodes see each other via mosmon but report the head
as DEAD, the opposite is true from the head.
I can run an sample program on any node and it will
migrate and balance to the other nodes just fine BUT I
need the head to accept and run the jobs and farm it
out to the nodes and that is what is not happening.
I have read everything I could find and have tried
using the -g 1 option to setpe as well as various
incarnations of OpenMosix.map.
The auto discovery DOES see the nodes but it assings a
variety of numbers to all of them but FAILS to migrate
or load balance ever. I have forced via modctl the
speed of the head to "1" and had no effect. I could
manually FORCE the migration by PID but I need an
automatic solution.
Also, the docs state that in a dual interface the ALIAS
option can be used but that only yield an error during
startup in my case.
Any ideas?
Logged In: YES
user_id=563899
this may be a shot in the dark, but on the "head" node did you set
AUTODISCIF to the correct interface in /etc/openmosix/openmosix.config
before starting the openmosix daemon via the initscript?
Logged In: YES
user_id=1196698
probably hostname of the head is not set to the internal
network (192.* but to the external legal one.
I set in hosts file local names and assigned them to all nodes.
It removed this type of problems. ALIAS in openmosix.map
seems to be not always working properly.
Tomasz
Logged In: YES
user_id=1021609
I had a similiar problem being cuased by my switch. I had to
turn off all IGMP features. You are getting an initail
broadcast acroos teh cluster and then at somepoint a filter
kicks in causing problems. Have you checked teh output of
dmesg? Also take a look at trouble the shooting section of
teh openmosix How-to.