(1) GES Strawman document

GES discussion
PGI working slides
stick to OGSA-BES
slide 5:
list of requirements
slide 7:
general remarks
OGSA-BES tuning issues
distr.way of classification of specs
Morris:
cannot agree on complete set of data staging
Andrew:
well scoped profile
Balazs:
add monitoring to conclusions
Andrew:
WSRF communities
BES factory port type
must support factory attrib.
Morris:
interop. scenario:
resource model partly supported
David:
slide 8:
GLUE describes everything -> info for GES endpoint
metadata attached?
info available through separate info service
Andrew:
in operation: info service, used for searching
should be moved to a separate discussion but the schema is very important
Morris:
GLUE uses some properties not mutual exclusive
preceise info
info to interconnect
Andrew:
GlUe2 madatory
what are the critical things you have to know for GLUE?
Morris:
not specify details here
Moreno:
define an appropriate subset of GLUE to describe factory attr. doc.
desc. BES endpoints use GLUE2
Andrew:
publish: not push in info service
Balazs:
return XML info
important functionality
Morris:
agreement?!
David:
GLUE is just schema
Morris:
render GLUE2 to XML
Balazs:
make info available for client
Andrew:
get attr. doc.
Morris:
cannot search
Moreno:
concern whole XML rendering -> maybe very large doc!!
not a good idea to return whole doc to client
Andrew:
get dynamic eprs and activities back
David:
having this info
practical point of view:
not how poeple work at the moment
requ. activities hav to get GLUE schema
Andrew:
highly dynamic, 500-600 jobs
metadata about container
Morris:
rather static info, less compared to dynamic one
interested in the static part which is small
Balazs:
should not dive to deep in details
David:
how to attach metadata to BES endpoint??
reason to separate ativity from status
Morris:
mapping 2nd version of BES or separate service
slide 10:
Balazs:
functional requ.
multi batch queues
request routing
share through single endpoint
12:
basically job operations for create activity 
location of staging directories should be communicated 
Andrew:
question:
if JSDL goes in -> epr back, stat not running
port type on activities itself
Balazs:
changing status of activity
not much to add
Andrew:
in BES group there have been long discussions about status changing
this was in the specs originally
Andrew has info (emails) to this
Moreno:
do have operations for suspend, ...
they are useful in general
Morris:
support data push paradigm
Moreno:
purge operation
different from terminate activity
slide 24:
remove all info from exec. service
examine output of activity
could implement by change status
David:
how do you handle if one fails in the middle of running?
do it like MPI -- kill all??
Balazs:
other discussion if multiple op. needed or not
Andrew:
info adding to BES garbage collection
Balazs:
in NorduGrid clients can request purging -> automatic garbage collection
Andrew:
it's a hint
Balazs:
yes
David:
if you don't get every info of single op. back to the epr -> have to kill everything
Andrew:
independant jobs once submitted
David:
as long as you submit them as independent JSDL jobs, then it should be there
user has to know all the endpoints
not method of restarting

due to bad quality -> closed, ongoing email discussion 


(2) AOB