Mills@udel.edu (11/13/90)
smb, The problems, says Van Jacobson, are mainly due to the fact that Sun has incorporated the System V timekeeping code in 4.x and that this code is broke to max. The preferred solution is to abolish the clock interrupt code and retrieve the corresponding code from older releases, which admittedly is ugly. Apparently, interrupts get turned off every thirty seconds or so while dirty pages are migrated to backing store, at least for diskful systems. There are other places where interrupts are masked off up to 150 ms. The eventual fix is either to resist the urge to mask off interrupts for that long or to provide some way for the hard-clock routine to make up for the lost interrupts in the interface with the soft-clock routines. I suspect fixes will be forthcoming somewhat soon in any case, since we need good clocks for meaningful DARTNET experiments. Dave