Aphs.190 net.unix-wizards utzoo!decvax!duke!phs!dennis Wed Jan 6 10:54:27 1982 swap algorithm change This is a reissue of an article which I mailed to unix-wizards@sri-unix, and which apparently got out onto the ARPAnet, but never got gatewayed back to USENET. I imagine you can catch it at sri. I have installed and have been running for several months a modification to the swap algorithm which seems to give better swap performance for systems with lots of terminals. A less thrashy swapping pattern is observed. The mod is to use p_size*p_time instead of just p_size when finding a large process to swap out. More specifically, when calculating maxsize, instead of comparing it to rp->p_size, compare it to (rp->p_size*rp->p_time), and set it to this also. p_size+p_time was tried, but it didn't work as effectively. The effect of this is to swap out processes that have been hanging around doing nothing, ie, inits and gettys. I've observed that they stay in until somebody runs spell or something similar and then they all go out at once. This is a large instantaneous load, but it certainly beats getting your shell swapped in and out repeatedly, which is how I noticed the problem. Watching gettys languish in core while somebody's current shell was swapped out annoyed me. I have installed this in both vanilla v7 and 2.8bsd, and it works fine in both cases. Our 11/44 has .75Mb and 48 terminals, usual user load around 15, and this has improved response time noticeably at certain times. It's certainly cut down on swapouts. ----------------------------------------------------------------- gopher://quux.org/ conversion by John Goerzen of http://communication.ucsd.edu/A-News/ This Usenet Oldnews Archive article may be copied and distributed freely, provided: 1. There is no money collected for the text(s) of the articles. 2. The following notice remains appended to each copy: The Usenet Oldnews Archive: Compilation Copyright (C) 1981, 1996 Bruce Jones, Henry Spencer, David Wiseman.