Lothar and Greg pointed out that the USCMS Test Grid has received some very positive publicity (both during CMS Week and the LHC Grid Week) and is currently in the "lime-light so-to-speak". This is indeed good news and reflects well on the hard work and effort that everyone has put into creating this grid. However, this also carries a heavy responsibility and we are thus at a critical time to demonstrate the success of the USCMS Test Grid! Status reports: Florida: -------- Ready for MOP slave operations--already demonstrated during CMS week. After the 200k MOP production, the kernal on testulix will have to be updated. Florida will take that opportunity to install the VDT 1.0 on testulix (and GDMP 3.0 alpha). For this production (CMSIM 200k events), Florida will continue with the old kernal and VDT 0.5. San Diego: ---------- Ready for MOP slave operations, but need to test the local MOP setup. GDMP needs to be looked at (possibly restarted) as the heart beat scripts have stopped transfering files between FNAL and UCSD (as of 17.03.2001 the heart beat scripts are working again :-) Caltech: -------- Also ready for MOP slave operations, but need to test the local MOP setup configuration. Wisconsin & Fermilab: --------------------- Almost ready for MOP master operations. The current situation is that the backup of MOP is OK (used for the demo). But there have been some new GDMP changes that need to be accounted for, such as the fact that GDMP is not currently launching because of problems loading shared object libs from the local platform. There are some site-specific work-arounds that Peter is investigating, so we should be prepared to give Peter any site specific information he requests. In addition, MOP and IMPALA need to be resynchronized--Greg and Peter are working on this. Finally, Peter is ensuring that that the grouping of the many mini-dags into one huge dag is robust. Indeed, this experience is providing very valuable information about setting up a testbed for useful work! Current MOP situation: ---------------------- There is a strong push to finish the 200k MOP assignment as soon as possible. However, in view of the tweeking needed for the current version of MOP/IMPALA/GDMP, Peter suggested that we aim for MOP operations on 22 Friday. This would allow Peter to understand the current problems and assure a robust MOP environment. In the mean time, the other three sites (San Diego, Caltech, and Florida) should strive to be as ready as possible. *** The next USCMS Test Grid Telecon will take place on *** *** Wednesday 20 March at 16:00 EST (13:00 PST) *** VDT 1.0: -------- Lothar and Greg pointed out that including GDMP 3.0 alpha 3 in the VDT would be more helpful to CMS than the currently included GDMP 2.1. Alain explained that GDMP 2.1 was included in the VDT because the EDG would not support GDMP 3.0 alpha at this time. However GDMP 3.0 is a major advance in functionality and it is a PPDG release. In addition, in the short term, the U.S. has Shahzad for GDMP 3.0 alpha 3 support and it appears, in any case, that GDMP 3.0 alpha will become supported in a matter of weeks. It was thus decided to write an email to the VDT developers, asking for a new VDT release which includes GDMP 3.0 alpha 3 (or higher), and that it be released as soon as possible. Alain Roy estimated that, given the "green light" this could be accomplished relatively quickly--a couple of weeks time :-) Rick and Dimitri have been testing the latest VDT 1.0 release with good results. Their conclusion is that all VDT parts (client, server, developer) install very easily (almost trivially) and congratulated Alain Roy and the VDT team. The steps needed to configure the individual components require more attention, however. This is not really intrinsic to the VDT, but rather intrinsic to the individual components (e.g. Condor itself can often be difficult to configure correctly). Alain said that this is an area where much more work needs to be done and he is currently thinking about different stratagies. Dimitri pointed out that one needs to be root to install the VDT (not mentioned on the VDT web site). Rick asked if it is proper to install Globus as root (Globus instructions indicate that Globus should be installed under user globus). No one really knew the answer to this. Rick had some confusion installing both the VDT client and the VDT server on the same machine (as the Condor and Condor-G executables have the same name). Alain and Peter said that Condor and Condor-G can co-exist peacefully, but that one needs to switch between the two (using a shell script, or something similar). Finally, Rick suggested that two or three canonical examples for VDT configuration would be very helpful (accounting for common firewall setups, shared file system setups, etc). We should document the exact configurations that each USCMS site uses to install and commision the VDT and send this to Alain. In the mean-time, Rick and Dimitri are compiling a wish/problem list for the VDT 1.0 release and will send it to Alain.