idallen@watmath.UUCP (10/11/84)
For CPU limits to be truly useful, processes must STOP, not die, when they exceed their limits. Once stopped, a mechanism for increasing the allowed CPU must be provided. Rather than changing all the existing code to incorporate the proper signal handler, can one change the kernel to STOP runaway processes, and also provide a program to let users increase the CPU allowed to a STOPped process? (If renice(8) can alter process priority, surely a similar mechanism can alter allowed CPU.) Having this facility would let us set a reasonable CPU limit on our student machine (to catch runaways) without having editors and other processes that really need the CPU die out (they would STOP, not die, and could be restarted with a bigger CPU limit). I am aware that setting a default SIGXCPU signal handler in crt0 in the C Library would allow new programs to STOP when they exceed their limit, but the mechanism for increasing the limit for a STOPped process is missing (short of arbitrarily increasing the limit by some magic amount), and programs that don't use the library don't benefit. A similar way of treating SIGXFSZ should also be invented. -- -IAN! (Ian! D. Allen) University of Waterloo