Il 28/08/2013 16:05, Benoît Canet ha scritto: >> Yes, I'm afraid that the burst is problematic here since the current >> > throttling algorithm doesn't help (and IIUC it behaves as if it had >> > max=avg/10 due to the 100ms slice time). This is Windows so no CFQ >> > here. Do you need to set bps_max=1 or will bps_max=0 work too? > In it's current state the code will overwrite the burst if it's set smaller > than > avg/10. > The result will be a small initial spike of size avg/10 then it will cap to > avg.
Can you whip up a patch to avoid that? Or to honor bps_max if it is specified (making avg/10 simply the default)? Paolo
