Present: Harvey Newman, Julian Bunn, Jim Amundson, Greg Graham, Dimitri Bourilkov, Rick Cavanaugh, Jorge Rodriguiz, Suchindra Kategari, Ian Fisk, Peter Courvares, Mike Wilde, Jenny Schopf 1. Software Base: a. VDT 1.0 Use of the VDT was considered. Greg asked about the content of the VDT: Condor 6.3.1, ClassAds 0.9, DAGMan 6.3.1, Globus 2.0, GDMP 2.x. Mike explained that the VDT will minimally package everything together so that all components are configured to work together. The foreseen timeline for VDT 1.0 is early to mid Janurary. As the timeline for the initial testbed is (loosely) based on the EAC review in early Januray, the descision was taken to build the initial testbed using as many of the VDT components as possible and configure them by hand. Further discussion ensued: the SC01 demos require Condor-G which is currently only released for Globus 1.1.3. Peter explained that a Globus 2.0 based Condor-G version is not foreseen before mid to late Janurary. Hence, the consensus is to initially use Globus 1.1.3 and then migrate to Globus 2.0 as soon as VDT 1.0 becomes available. This choice requires that GDMP 2.0 alpha be used (as it is based on Globus 1.1.3). The base list of software to be installed and configured at each site is then agreed to be: Condor 6.3.1 Condor-G 6.3.1 DAGMan 6.3.1 ClassAds 0.9 Globus 1.1.3 GDMP 2.0 alpha Objectivity 6.1 b. CMS software environment It was decided not to install a "full build" version of the CMS software at each site. Rather versions of DAR files (tarballs containing the CMS executables, shared libs, environment shell scripts, etc) will be installed at each site. To start things off, the DAR file for the "Big Minbias" production (already in production and being used) will be installed at each site. This includes CMSIM 123 and ORCA 5.3.1. The possibility of installing a CMS DAR file "on-the-fly" will be investigated as one of the Testbed goals. 2. Operating System: Everyone agreed to use "vanilla" RedHat 6.2 3. Interoperability with the US-ATLAS grid and EDG. Jenny mentioned that US-ATLAS and the EDG currently require that each site run AFS. As the CMS software does not impose this requirement, and to keep things as simple as possible, it was decided not to require AFS at each site. However, as eventual integration with the US-ATLAS and the EDG testbeds will likely happen, we should keep this overhead in mind. 4. Machines at each site: a. Caltech: a few nodes. Julian wondered how best to partition an existing cluster into two clusters. Jorge described the situation in Florida where a test-cluster was partioned out of a larger cluster by creating another distinct head-node for the test cluster. (This is part of an on-going study of setting up multiple "head" nodes and then re-partioning the cluster and re-installing the OS on any selected partion using Rocks.) Harvey brought up the issue of purchasing nodes: which processors should we buy (PIII will soon be out of date, PIVs are expensive, Athalons do not appear to be fully tested)? This was realized as an imporant topic, but no recommendations were made. The discussion will continue via the new USCMS testbed mailing list. b. Fermilab: 3 nodes immediately, more nodes will become available in the future. c. Florida: 1 + 4 nodes (1 public head node + 4 private worker nodes) d. San Diego: 1 + 4 nodes. e. Wisconsin: several nodes. Peter indicated that these nodes will be in the Condor Pool at Wisc. The number of nodes can increase if the need arises. 5. Account management This was recognized as a difficult (and general) problem and we "punted" concerning the scalability issue. For now, it was decided that each user must individually request an account from the contact at each grid site. Group accounts will not be accepted. In addition, the user must have *already* received a Globus certificate and must include his/her "Distinguished Name grid-mapfile entry" in the email request. The site contacts are: Caltech: Julian Fermilab: Jim Florida: Rick San Diego: Ian Wisconsin: Peter The site contacts will be responsible for "walking" the user through the application process. The user must then fill and return any required paper-work as requested by the individual grid sites. This initial solution to account management will be reviewed as the testbed matures. Instructions for new users will be placed on the USCMS Testbed web pages. 6. CA issues and Security were not discussed. The current mode of operation is to use - Globus, initially - ESnet, when available (early 2002) Fermilab requires Kerberos authentication and this should be discussed in the next telecon on 17 December. 7. Timeline a. It was agreed that we will aim for an operational testbed in ~2 weeks time. b. The next telecon will be at 16:00 (Eastern Time) on Monday 17 December. Appendix: Potential users of the USCMS Testbed (union of all CMS-PPDG and CMS-GriPhyN members--is anyone missing?): Harvey Newman Julian Bunn Koen Holtman Conrad Steenberg Takako Hickey Iosif Legrand Vladimir Litvin Suresh Singh Edwin Soedermadji Lothar Bauerdick Ruth Porder Jim Amundson Greg Graham Vivian O'Dell Shazhad Muzaffar Paul Avery Dimitri Bourilkov Rick Cavanaugh Holger Stoeck Jorge Rodriquez Suchindra Kategari Ian Fisk Miron Livny Alan Roy Peter Couvares Raj Rajamani Mike Wilde Jens Voeckler James Patton Asad Samar James Branson