The cluster/hitfinder represents most of the cpu usage for reconstructing particles in the TPC [12]. Here, we present some numerical examples of the cpu cost for TCL in current form. Running the slow simulator and TCL for a subset of the TPC, and for the full TPC indicates that the cpu time scales with the number of sectors considered, as one would naively guess.
Numbers are presented for the software running on two machines; the first numbers are for rsgi00.rhic.bnl.gov, an SGI machine with a 40 MHz clock, and the numbers in parentheses are for rncsv1.lbl.gov, a Sparc-20 equivalent with a 60 MHz clock. A benchmark test has clocked rncsv1 as about 3 times faster for integer operations [13]. For a Au+Au event (generated by HIJING), the clusterfinding (TCL module) takes 2.7 (1.0) cpu sec/sector, or 64.8 (24.0) seconds for the full TPC. With standard switch settings (see Section 5), the hitfinder (TPH module) takes 61.8 (55.5) sec/sector, or 24.7 (22.2) minutes for the full TPC. Running TPH with no deconvolution (TCLPAR.DECON=0) takes 23.5 (11.2) seconds/sector, or 9.4 (4.5) minutes for the full TPC.