Rene Bartsch wrote:
You don't need to report the time the kdvb-fe threads but the remaining idle time, this should always be >0.0%Am Die, 2003-06-10 um 15.23 schrieb Holger Waechtler:Rene Bartsch wrote:Shouldn't there be a fail-safe solution limiting CPU-load of kdvb-fe? I don't think it is the right way to break a machine which may have other important tasks because of bad weather ...have you ever experienced something like this?At my parents while hard rain and lightning the signal is sometimes lost completely (90 cm dish directed and adjusted to ASTRA). Just a few minutes ago two of three kdvb-fes on a server I'm maintaining suddenly jumped up to 40 % each for some seconds. Usually they're between 0.1 - 1%.
20% of what? of a machine running idle or one that's running 10x overloaded? Right now the code is written so that the threads should only eat cpu idle time. When no idle time is left then they'll zigzag much slower...the driver is trying to (re-?)aquiring the signal by zig-zag-scanning the frequency range around the center frequency. This is used to compensate temperature drift of LNBs and RF components used in the transmitters and receiver units.I also do not understand for what all this additional CPU-load is used.
Ok, but shouldn't be there some kind of limit? Think about four cards in a machine starting zig-zag scanning because of some lightnings and the machine acts like slow motion. Some limit in the driver like all kdvb-processes together mustn't have more than 20% CPU-load would be fine.