Choose timezone
Your profile timezone:
TCG Meeting
Description
A phone conference has been set up:
To dial in to the conference call:
Call: +41227676000 (Main)
Enter access code: 0170047.
To join the web conference, click here: https://audioconf.cern.ch/call/0170047.
To dial in to the conference call:
Call: +41227676000 (Main)
Enter access code: 0170047.
To join the web conference, click here: https://audioconf.cern.ch/call/0170047.
TCG Minutes 26/03/08
Present - Oliver Keeble (chair), Nick Thackray, Markus Schulz, Xavier
Jeannin, Claudio Grandi, Cal Loomis, Andrei Tsaregorodtsev, Jeff
Templon, Stefano Belforte.
Minutes of last meeting and action list;
#6624 - approved by JT. CG to check with CW that this is in the doc.
#6623 - talked to Flavia and emailed the two, done
#6621 - to be discussed in Bologna
#6620 - no patch. Massimo trying to understand how to do this best,
currently have the situation that CREAM needs voms 1.8 which has not yet
been certified.
#6619 - OK has put the relevant people in touch. Imperial College,
although not in SA3 in EGEE-III, will be doing blah/SGE integration and
is in contact with the relevant members of the CREAM team.
#6338 - JT - tried to get feedback and received nothing. TCG decided not
to implement the (will be revisited once star has been implemented).
Ultimately will be address by auth framework. Action should be closed.
#6336 - No progress
#6337 - sent to Erwin a list of things to achieve before end of project.
Keep task open until approved.
#5952 - can we certify DM stuff for python 2.3-2.5?
MS - we need a policy in the project about whether we stick to reference
versions otherwise things are unmanageable. How do we prioritise? JT -
compromise; it's supported for a ref version, however project policy is
try to be as compatible as possible. If you have a fix, send it and
we'll try to include it. This will make it easier for us when we migrate
reference versions anyway. CG - they can be submitted as enhancements.
Then we can track.
Decision ; We will accept enhancement bugs on components which are not
working with python versions outside the reference version (2.3) but the
solution will only be implemented if the binding can be made more
general (ie applicable 2.3-2.5) and no separate packages for eg python
2.5 will be released.
#5415 - MS - how can a sysadmin check that they support properly a VO
without being a member of that VO. JT - can an admin launch SAM tests to
non-production sites via any VO. To be authorised to do this, he has to
be in dteam/ops. Requirement on SAM. Bug to go to SAM and link to task.
bug #34959 has been submitted.
#5401 - this remains open. JT- Dennis (SA3/Nikhef) has looked for
someone but noone has picked it up. MS - cream/blah will come with LSF
support. JT - can we say in this task that for blah/lsf it's the blah
developers who are responsible? CG - not really. MS - INFN dropped this
so they could take over running experimental WMS services. Still need
accounting and infosys. Closure of this task to await clarity in the
SA3/EGEE-III execution plan.
Requirements Follow Up
======================
OLIVER: Any further requirements to input?
There were no comments. Experiment attendance was low (at the time we
didn't realise that Stefano had joined but was inaudible...) so decided
to start with the NA4 requirements.
> 502 Membership with multiple organisations must work correctly:
No further comments. Remains open until VOMS deployment is complete.
CAL: Not all services use VOMS information.
MARKUS: Which services are you referring to specifically?
There was general discussion on which services use VOMS properly.
JEFF: Has anyone tested being in 2 VOs at a time?
There was some confusion on exactly what Jeff was asking. Eventually it was agreed that this hadn't been done yet.
OLIVER: This requirement (502) is very general; maybe we should ask for it to be made more specific.
CLAUDIO: Is 502 the same as 101?
All agreed it is a special case of 101.
All agreed to put 502 in with 101.
> 101 JEFF: Would like to ask that all VOs should use VOMS for production or software installation jobs:
As no experiments were representation, will defer this discussion.
Short Deadline Jobs
> 511 The release should support SDJ at the level of the batch systems:
There is a working group for this. Experience shows that general solutions for configuration of batch systems doesn't work as sites are very specific on how they configure.
Documentation exists for this.
ACTION on SA1: Ask ROCS how many sites would like information on configuring batch systems for short deadline jobs.
> 512 The resource broker has to be able to identify resources that support SDJs:
[Cal to send number of the bug tracking this one]
> 513 SDJ resource access should be controlled via ACLs:
It was agreed that this should be merged with 101.
> 514 Modify system to ensure shortest possible latency for SDJs:
NA4 stated that this should stay on the list but that it is lower priority.
> 521 Use a batch system that can handle the "CPU count problem":
Agreed that this is fixed in the MPI configuration we have.
This can be closed.
> 521 Publication of the maximum number of CPUs that can be used by a single job:
Possible with GLUE 1.3 but not systematically done in production.
The GLUE parameter should be set to a default value of 1 by YAIM.
OLIVER: Please submit a bug for YAIM.
Requirement will stay until Cal is happy with correct publication by majority of sites.
> 522 Publication of whether the home directories are shared (alternatively transparently move sandboxes to all allocated nodes)
The current solution is working so this can be closed.
> 523 Ability to run code before/after the job wrapper invokes "mpirun":
Current solution works. This can be closed even though the suggested implementation (WMS prologue/epilogue) is not used.
CAL: It would be nice for prologue and epilogue to also work.
Disk Space Specification
> 531 Specification of required shared disk space:
OLIVER: Will CREAM CE fulfil this for the batch systems it supports?
CLAUDIO: Yes but only if the correct BLAH plug-in is written by someone - most don't currently exist
CAL: Does current WMS send info to CE now?
CLAUDIO: No because there is only the LCG CEs in production.
Conclusion: Need deployment of CREAM CE + plug-ins for this to be resolved. Not expected any time soon.
> 532 Specification of required local scratch disk space:
As for 531.
ACTION on Markus to check if it is possible to distinguish local storage/distributed storage through the GLUE schema.
If not then ACTION on SA3 to request it.
Publication of software availability and location
> 541 Publication of the Java and Python version; Mechanism to find the required versions of those packages:
This is in GLUE 1.3 but not read in by WMS.
CAL: Is the list of GLUE parameters that are loaded into the ISM hard coded or controlled by configuration parameters?
CLAUDIO: Not sure; will check.
ACTION on JRA1 to check if WMS will correctly match using the published information.
ACTION on NA4 to check how info providers would work
ACTION on Oliver to find out from Steve T. if he is taking the above into account for his work on correct publication of sub-clusters.
======================================================================
Minute taker's laptop dies, responsibility transfered, prepare for a
change of style.
======================================================================
551: CG - Job priority can steer jobs to queues with higher priorities,
but single jobs can't hold the information of specific priorities, Even
the new authorisation system will not allow this on a job by job basis
552 JT - can be seen as satisfied in a limited sense since groups can
have different priorities. The VO control bit is not satisfied
500 category remains open.
561: OK - DAG is not currently working on the SL4 WMS (collections are
fine).
CG - This has to be fixed
Short discussion that we can't stop the release for this.
562: Manipulation of job states for compound jobs.
Claudio: foreseen, has to be verified
563: covered by the same.
All 56X will be covered by the DAG implementation.
570: CL - Covered by AMGA which is now in the release.
580 Will be all addressed by Hydra, the requirements will stay
590: Software licenses
JT - Looks like the software version.
CL - No this is very different, the job needs to be get the licence key
at run time.
JT - This is very difficult,
CL - This is difficult, the requirement should stay on low priority.
MS - Why not ask the makers of Flex to cooperate on gis security
CL - Already done, but need to get the schedulers. 592 ( the second one)
JT - For some user groups this is not low priority, it keeps some user
groups out. Especially small VOs are kept out.
CL - For the two groups I am aware the Flex solution works well, Jeff
and Cal will discuss this for Jeff's users offline.
600 Database access:
All still valid and very much needed. We will start to setup a data
base working group.
MS - we should get the Coral developers to contribute.
CL -The mandate will presented to the TCG as soon as it is ready
(approximate 1 months from now).
701:
Output data support in the JDL
Claudio: Attribute that was in the JDL for some time, but not used. Has
been removed, but is now wanted. Can be done easily, but not clear how
important.
OK - Does anyone want to support this requirement?
CL - I need to understand this better and so we better leave it.
CG - I had added it by hand.
OK - NA4 items finished. Please provide the necessary assessment for the
other items so we can proceed with the list at the next meeting. Next
meeting with Erwin, 9th April.
AOB - Feedback from CG on issues already discussed;
CG - Feedback from the developers. 31278 (WMS should prevent non-SDJ
jobs from being scheduled on SDJ CEs) is fixed.
CG - More feedback on software versions published and match maker. The
only one in use is the software environment field. New filters can be
added quickly as soon as it is known which class it is.
There are minutes attached to this event.
Show them.