E866 Off-line Analysis Plan The off-line analysis plan consists of a number of different levels addressing various needs and with different facilities. Database - A common database that will serve analysis at all stages outlined below, including wide-area network and programmed access, will be used to collect and disseminate geometrical alignment data and other detector and run data as well as to keep track of all analysis done on each data sample. Backend Analysis - Using approximatly 150 SpecMarks of SGI UNIX workstations at the portakamp, which will be supplied by Fermilab, we will analyze data in real time. Data will be shipped from the on-line data acquisition system via ethernet and various monitoring processes will verify it's integrity including full analysis and tracking of up to 10% of the events. The analysis will address issues ranging from raw detector histograms and dead channels, to overall efficiency for detecting J/Psi events (with low statistics). Bicycle Online - UNIX farms provided by Fermilab will be used to analyze about 20% of the data soon after it is taken. This analysis will check the readability of the raw 8mm data tapes, and will be used to get initial results for cross sections and ratios. Here we will also to investigate possible rate-dependent effects on the cross sections and ratio. This analysis will also serve as the prototype for the final production analysis, which is expected to begin soon after data taking is complete. Production Pass 1 - The same UNIX farms used for the bicycle online will be used for the production first-pass analysis of the complete data sample. This production analysis will include the most compute intensive part of the the offline analysis. This will include track reconstruction in the drift chambers, traceback to the target region through the SM12 magnetic field, reconstruction of basic physics quantities, and writing out the results for events that pass to a data summary tape (DST). A UNIX farm of about the same power as was used in the production analysis of the E789 data should be adequate for this job and for the bicycle online above. This system was a set of four farm systems with six IBM compute nodes each. We estimate that such a system can analyze a E866 8mm tape of raw data in about 4 hours. With the expected number of tapes being about 1000 the full analysis should take about six months. Previously the 24 compute nodes of the farm were divided into four smaller farm systems. We hope that with improvements in the way input and output data flow are handled all the farm compute nodes can be in one, or at most two, pieces. This will ease the management and bookeeping of the production process. Physics Pass 2 - The output of the first pass will be distributed throughout the collaboration for final analysis to physics results. UNIX computing resources at Los Alamos, the Fermilab Physics Department, and at all the other collaborating institutions will be used for this. This type of analysis will also be applied to the output of the bicycle online analysis during the data taking. Also during data taking some of the institions computing resources at large may be located at Fermilab to provide easy access for the collaborators who are on site taking shifts. In order to support the physics analysis off-site, we will rely on our VME based tape copy facility to make copies of the DST tapes for distribution by Express mail. For smaller sets of data we will also rely upon distribution over the network. Reliable network connections will also be important for distribution of programs and databases.