120 likes | 226 Views
GLUE Schema - Status Report DataTAG WP4 meeting Bologna, 29 July 2003. Sergio Andreozzi INFN-CNAF andreozzi@cnaf.infn.it. OUTLINE. The Information Model – open issues Computing resources Storage resources Monitoring extensions The LDAP Implementation for MDS Integration status.
E N D
GLUE Schema - Status Report DataTAG WP4 meeting Bologna, 29 July 2003 Sergio Andreozzi INFN-CNAF andreozzi@cnaf.infn.it
OUTLINE • The Information Model – open issues • Computing resources • Storage resources • Monitoring extensions • The LDAP Implementation for MDS • Integration status
Computing resources Issue 1: CE viewpoint of the cluster • The subcluster concept is suitable to provide a summary description of the nature of a cluster in term of host characteristics • It does not properly suits brokering needs • This leads to the current EDG constraint of one subcluster per cluster • This is a limitation because: • Does not accommodate the hypothesis of heterogeneous clusters • Does not accommodate flexibility in batch system configuration • MISSING TO EXPRESS DESCRIPTION OF NODES ASSIGNED TO A SPECIFIC QUEUE
Computing resources Issue 1: CE viewpoint of the cluster In this cluster, a computing element is configured with a node assigned to CMS VO. A specific software is installed in that node. infoService gatekeeper Batch server Head node … CPU:PIV RAM:2GB OS:Linux CMS CPU:PIV RAM:2GB OS:Linux ATLAS CPU:PIII RAM:0.5GB OS:Linux CPU:PIII RAM:0.5GB OS:Linux
Computing resources Issue 1: CE viewpoint of the cluster In this cluster, a computing element is configured with a node assigned to ATLAS VO. A specific software is installed in that node. infoService gatekeeper Batch server Head node … CPU:PIV RAM:2GB OS:Linux CMS CPU:PIV RAM:2GB OS:Linux ATLAS CPU:PIII RAM:0.5GB OS:Linux CPU:PIII RAM:0.5GB OS:Linux
Computing resources Issue 1: CE viewpoint of the cluster this association needs to be introduced and described
Computing resources Issue 2: multiple entry points to a queue • The CE Unique ID contains the contact point • For scalability issues, in big cluster it is envisioned: • To separate gatekeeper from batch server • To deploy more than one gatekeeper • This leads to the duplication of CE entities • Needs to refine the CE definition • Possible solution: • refine the CE ID, (informally) from the grid entry point into a queue of a batch system, to the queue of a batch system • Entry points are properties of the CE
gatekeeper gatekeeper gatekeeper … Access node1 Access node2 Access noden Batch server Head node gris Can run on an access node, on the Head node or on another machine … Worker node Worker node Worker node Worker node queue Computing resources Issue 2: multiple entry points to a queue e.g.: from CEUniqueID: accessnode1:2119/jobmanager-pbs-red accessnoden:2119/jobmanager-pbs-red to CEUniqueID: headnode/jobmanager-pbs-red (the list of accesspoints are attributes)
Storage resources • The Grid Resource is the space being shared • Glue SE Schema aims at provide a generalization for several Storage Resource Manager • Glue SE Schema 1.1 is mainly based on • experience on using disk-servers with a set of protocol to access/transfer data • design documents related to SRM
Storage resources • Current deployment experience within EDG 2.0 (where SE is the EDG-SE) is providing useful feedback • Several refinements are needed in order to deal with new use cases • I’m working on a proposal with use cases; this will take into account most of the features discussed during last CHEP • (ref: http://www.cnaf.infn.it/~sergio/datatag/glue/v11/SE/GlueSE_revision.html)
GLUE Schema extended for monitoring • within DataTAG we have extended the host entity with several metrics • these have not been proposed for inclusion in the GLUE schema • any action on this?
GLUE Schema integration in VDT • Early July, meeting with Alain Roy at CNAF • Achievement: • By this week in VDT 1.1.10 there will be: • Updated GLUE Schema, extended version • EDG CE Info provider properly configured for CONDOR LRMS