Interfacing the CMS BOSS system with the EDG Relational Grid Monitoring Architecture
3.4.3
   
Summary

The CMS batch job submission system (BOSS) was developed to provide monitoring and bookkeeping of jobs submitted to a compute farm. BOSS itself is not a batch system and it interfaces to a local scheduler (PBS, Condor etc.) through a set of scripts. BOSS extracts from a specific job the information to be monitored and stores it in a database. The current release of BOSS (V3.3) uses MySQL as the database system.

The goal of this task was to provide EDG middleware enabled access to the BOSS database. For this task we use the EDG Relational Grid Monitoring Architecture (R-GMA) developed by WP3. The main concept of R-GMA is for a source of information (a producer) to subscribe to a registry. Potential users of this information (consumers) then lookup the registry. Having located a source of information the consumer connects directly to the producer and data is streamed to them.

Our solution consists of an R-GMA producer has been integrated into the BOSS job wrapper (C++ code), and a separate “receiver” that consumes information from R-GMA and updates the local BOSS database (Java code). Data is sent through R-GMA in the form of messages describing the updates to be made rather than complete tuples. Although this introduces the risk that the BOSS database will be left in an inconsistent state if a message gets lost, it insulates the R-GMA transport layer from any changes in the BOSS database structure, such as the addition of new job types. (The BOSS database is in any case validated against the correct data returned as a journal file in the output sandbox).

EDG Version Our software has been developed on version 2.2.4-12 of the EDG WP3 R-GMA software.
Brunel GRIDPP
User Documentation
Architecture
Technical
GRIDPP Project