Hi Arnie,
Sounds like you need to change some multicast IPs. All the nodes that you
want to appear in a single cluster should have the same multicast IP.
Despite your best efforts to explain it, I think you're probably the best
person to determine how you want your grid layout to look. :)
Also, gmetad works by polling individual monitoring cores - the cores
themselves don't provide any special treatment to the connecting
metadaemon. [... yet.]
Remember that gmetad is separate from gmond. You'll have to give hendrix
gmond a unique multicast IP to send its metrics on and then configure
gmetad to connect to localhost:8649 in order to get hendrix's info (as a
"cluster" of one box).
The web front-end doesn't really make any particular attempts to accomodate
singleton servers. Ganglia's underlying architecture isn't really
optimized for it but I do see the allure of having everything in one place...
Oh well. Hope this helps. Maybe we should make a bigger deal of what good
changing the multicast IP can do in the docs...
Arnie Miles wrote:
Here's the configuration:
One CISCO gigabit Ethernet switch. Plugged into this switch is
zappa.georgetown.edu, which is the master node of a 16 node OSCAR
Beowulf cluster, hendrix.arc.georgetown.edu, which is a stand-alone
computer (that also answers to time.arc.georgetown.edu), and
john.arc.georgetown.edu, paul.arc.georgetown.edu,
george.arc.georgetown.edu and ringo.arc.georgetown.edu, which are all
stand-alone nodes that happen to share a filesystem and use PVM, but
they are NOT a classic Beowulf cluster. I have other machines elsewhere
on campus all running gmonds.
Here's the problem: hendrix is my only gmetad installation (you can see
it at www.guppi.georgetown.edu), and it is supposed to report on all
clusters and all stand-alone machines on my 'grid.' Clusters report
their Master node on the top-level, and you can tunnel down to see the
compute nodes. This has been working just fine up until now.
When I tried to add john, paul, george and ringo, and have them send
their data to hendrix to be displayed, each of them will speak for
themselves, as well as for each other, AND hendrix (or time)!! For
example, john (which is a dual-processor machine) will report that it
has 5 nodes and 10 processors, and if you tunnel down on john you'll
find john (again), george, paul, and ringo, as well as either hendrix or
time. Additionally, hendrix will not report for itself. I've tried
making each of these machines 'deaf' in the config file, but then they
disappear entirely.
What am I doing wrong? Things have been going so well with ganglia here
up until now, but I don't see anything in the docs to help me. How can
I make each of these machines ignore the others and just report their
local information to hendrix?? At first I thought it was an issue with
sharing the switch, but if it were zappa would be involved. Then I
thought it was a PVM thing, but then why is hendrix involved???
Arnie Miles
Georgetown University
[EMAIL PROTECTED]
-------------------------------------------------------
This SF.net email is sponsored by: ValueWeb: Dedicated Hosting for just
$79/mo with 500 GB of bandwidth! No other company gives more support or
power for your dedicated server
http://click.atdmt.com/AFF/go/sdnxxaff00300020aff/direct/01/
_______________________________________________
Ganglia-general mailing list
Ganglia-general@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/ganglia-general
