DAMED-wg telecon notes from Wednesday April 24, 2002

Attending:
Martin Swany, Brian Tierney, Jenny Schopf, James Magowan, George Brett, Steve Englehart, Rich Tascano, Dan Gunter, Steve Fisher, Mary Thomas

Things that will happen before the next call:
Martin volunteered to write up the 4-5 ways that people looks at traceroute data, and will also talk with Beth about this
Dan will add agreed on additions to web page, and people will comment on it from there
James is going to draft what should be in a name versus what should be associated with it
Martin will draft the one page "what is a target"

Next call is one of:
Weds 8th, 10am
thursday 9th, 10am

-----------------------------------------------------------------------------------------

Jenny (and Brian) did a brief over view of where we sit in the world of other related groups, especially the
pg-monitoring group http://www.mcs.anl.gov/~jms/pg-monitoring/
glue-schema group http://www.hicb.org/glue/glue-schema/schema.htm

fredrica's workshop- related to grads documents as well

We (damed) are a real working group, although this is not reflected on the grid forum web page, we'll try to get this addressed

http://www-unix.gridforum.org/mail_archive/damed-wg/msg00037.html

Beth has an issue with say, traceroute, if each event has to be single value, need one event for each hop, how will they be bound together if you can't have multiple values

Suggestion - Dan could try to add traceroute to his example
Suggestion - Need input from Beth Plale, since other folks don't understand all of these issues
Several folks suggested possibly structuralize ways to map this data
Martin volunteered to write up the 4-5 ways that people looks at this data, he will also talk with Beth

Look at host processor load-

when there are different ways to gather the info, should it be host.processor.load.active or host.process.load.uptime

[there was a fairly convoluted discussion about what was an attribute here]

Brian asked how would we represent accuracy information here?
        thought was a subclass
Do we need to define units?
        after checking the charter, the answer is yes.
        martin noted that there are difficulties defining floats in both snmp and ldap - so maybe we should try to avoid them?
        problem - smallest unit to report would be needed to be defined now, how can you tell what this is?

Mary Thomas - would be nice to be able to define the units on the fly
        we had decided at GGF that if you wanted to report other than standard units you would define a different event type - so for example host.process.load.uptime.weirdunit

What's missing from top n?
        Extend top n to be representative of importance as well as type
        packet loss
        available mem (this may be useless because it's un-measurable....)
                martin talked about how this was measured in grads, it's used a lot there
        queue length
        service status (?)- can we generalize this?  maybe not leave this off for now? or have same status as for application?
                proposal - don't include, not widely used
        what about discovery stuff-
                architecture
                os type
                physical mem
                disk size
                host name (isn't this already the target)


Who's going to do these (add them to the table)?
        easiest of dan adds them, then they get modified

James is going to draft what should be in a name versus what should be associated with it

Next step should possibly be "opaque" target
        Martin will draft the one page "what is a target"

Next call - sanity check both of these things, and then decide what mapping should move to?

Next step - map to gos, and then xml, ldap, sql?