preece@uicsl.UUCP (02/23/84)
#N:uicsl:12500022:000:1329 uicsl!preece Feb 22 13:57:00 1984 Some questions on the 4.1 system sort program: 1. sort uses very little memory (although it is very careful to check that it can get the amount it asks for). This may date from pdp-11 days or it might be intentional. Does anyone know? In a sort/merge framework there is a tradeoff between the cost of disk access and the exponential behavior of the in-memory sort; does the memory size limit reflect a careful empirical determination of the optimal point on that tradeoff curve, or is it an artifact? On our 780 the sort runs faster if it uses more memory. 2. Has anyone done such experimentation and determined good values for the parameters in the program? The number of items that will fit in memory is derived by formula from the amount of memory, so just increasing the obvious values may not be the best way to go. The space for file buffers during the merge phase is involved, too. 3. Does anyone have alternative sort programs for unix? I hacked multi-line records into the existing version, but not too thoroughly; I'd be interested in a program with that capability and with better tailoring of internal parameters to the actual data, instead of just rigid guesses. 4. Is the sort that comes with 4.2 the same as the one in 4.1? (Likewise SysV)? scott preece ihnp4!uiucdcs!uicsl!preece