CERN/RD45/99/11
18 June, 1999
Interim Status Report to the LCB
Prepared as Background to the LCB Meeting of June 22nd 1999
This report provides a brief overview of the current status of the RD45 project, lists the short-term goals of the project (until the next LCB review) and outlines possible future directions. It has been produced as background for the interim status report to the LCB to be presented on June 22nd 1999.
The current production version of Objectivity/DB at CERN is V5.1. This version, which is fully interoperable on all platforms, is available for DEC, HP, IBM, Linux, SGI, Sun and NT. A bug-fix release, V5.1.2, is currently being installed and tested. As this version offers no major enhancements, there are no plans to migrate to it in the near future. V5.1.3 is also expected shortly – this version will support Microsoft Visual C++ V6.0 and the Sun CC compiler version 5.
V5.2 of Objectivity/DB, which will include a number of important new features of particular interest to the HEP community, is scheduled for release on all platforms by the end of June 1999. Although it appears unlikely that this date will actually be met, we can currently plan on receiving and testing V5.2 early enough for production use in 2000, as part of an official release of LHC++. Amongst the new features in V5.2 are the production version of the HPSS interface and numerous other enhancements designed by Andy Hanuschevsky / SLAC. The main enhancement that will not be satisfied by this release is support for federations much larger than 1PB (VLDB support). Assuming a maximum file (database) size in the 10 – 20GB range, the current architecture permits federations of some 500TB – 1PB. We are currently discussing with both SLAC and Objectivity as to how this restriction could best be removed, allowing federations up to ~100PB in size, on an appropriate timescale.
We have now managed to reach agreement with Objectivity concerning the Linux version of Objectivity/DB. The existing contract has been extended to cover the Linux operating system without other changes. That is, the 200 concurrent (floating) C++ full-developer licenses and 25 concurrent full-developer Java licenses may now be used on Linux systems, in addition to the previously supported platforms. This is true both for systems at CERN and outside, with the single restriction that usage is limited to work on the CERN Research Programme. As part of these negotiations, Objectivity have confirmed in writing that the enhancements that have been scheduled for the V5.2 release will all be made available free of charge and under the existing contract.
As agreed at the last major LCB review, RD45 has produced a series of topical "white papers" on a wide range of issues, including recommendations for the configuration of production federations, requirements for specific new features as well as a number devoted to the on-going risk analysis and possible alternatives. These white papers may be viewed, in both HTML and MS Word format, via http://wwwinfo.cern.ch/asd/rd45/recommendations.htm.
Additional white papers (such as this document), will be produced as appropriate.
The main issues that will be addressed in the immediate future relate to the milestones and recommendations set at the last LCB review. In particular, a status report will be prepared in time for the LCB Workshop in Marseilles, under the assumption that a review will be made at the time of the LCB meeting at CERN in October. The main focus of the RD45 workshop that is scheduled for July 12 – 15 at CERN will be on the completion of the work related to these milestones and the production of the status report.
Given the LHCC milestone for ATLAS and CMS for a choice of production "ODBMS" system in the second half of 2000, it is our assumption that future work should be related to the preparation of requirements and evaluation criteria, together with the identification of potential candidates, for this selection. It is currently planned that an updated list of requirements be produced on the timescale of early 2000. It would be based on the initial list developed at the start of the RD45 project, the milestones given to the project over the years, the production experience that has been gained, as well as the input from other projects such as MONARC and the LHC (and other) experiments themselves.
In addition to this work, it is felt to be essential that an extension of the risk analysis – together with continued support for existing users of the current services – be provided.
That is, users should be assured of continued support of the existing solutions, namely Objectivity/DB, until at least the end of 2001, or longer, depending on the decision taken at that time. In parallel, the key items identified as part of the risk analysis, such as the investigations into alternative solutions, should be given high priority. The possible solutions that are currently under investigation are discussed in more detail below.
The currently preferred solution remains Objectivity/DB and HPSS. However, in addition to the work on obtaining enhancements capable of meeting HEP requirements – such as VLDB support – it is clear that "fallback solutions" need to be considered. These are needed to protect against the case that Objectivity – for whatever reason – is no longer willing or able to satisfy HEP demands. However, it is important to stress that close collaboration with Objectivity needs to continue for the foreseeable future.
Although it is believed that existing RDBMS products are not currently capable of handling the full volume of LHC event data, a better understanding of the issues related to the potential usage of such a product is required. The existence of "ODMG wrappers", such as those developed by INRIA, could permit a "HepODBMS interface" on top of say ORACLE. Could such a solution provide the scalability and performance that is required on the appropriate time scale? Discussions with both ORACLE and INRIA are planned that will enable us to understand whether this direction is worthy of further study.
Given the vast experience that has been gained in ODBMS technology in recent years and the mature C++ environment that is currently available, it now appears that a "home grown" solution could indeed be feasible. Initial discussions both within the HEP community and beyond suggest that the manpower to design and develop such a system could be found, perhaps via joint projects with other high-end users (ESO and others), database research groups in various Universities, and perhaps also INRIA.
We believe that the work on the current milestones and recommendations is on track and that a status report can be prepared in time for the LCB workshop in September. Beyond this date, we feel that the main priorities, in addition to continued production services, will be related to the selection of an "ODBMS", currently foreseen for the end of 2001. As such, we believe that the on-going "risk analysis" work, including the identification and, if required, provision of alternatives, together with the preparation of the requirements and evaluation criteria for the above-mentioned choice, to be of high priority and suggest that these issues form the basis of RD45’s future activities.