On Wed, May 02, 2007 at 03:44:30PM +0100, Greg Kochanski wrote: > Mattia Dongili wrote: > >On Wed, May 02, 2007 at 09:36:27AM +0100, Greg Kochanski wrote: > >... > >>Suggestions: > >>1) allow sampling_rate_min to be set to smaller values, like 0.1 second. > >> The default sampling_rate_min should probably be 0.25 seconds. > >> If that were the case, the CPU would power up before > >> a window drag (or similar motion) was finished, and the response > >> would > >> probably seem acceptably snappy. > > > >actually the sampling rate is dependent on the transition latency for > >the processor. > > Can you explain further?
assume your processor takes 100ms to switch from one frequency to another. Ondemand takes this value as a base to calculate the polling frequency (min, max et al.) and to avoid the possibility to clog the system with to many transitions it sets a lower value that is a multiple of the transition latency. > >>2) Make the default scaling_min_freq be somewhat larger, perhaps 670 MHz > >>or > >> 20% of the max clock speed. This will (of course) reduce laptop > >> life, > >> but I suspect that it won't make a noticeable change. > > > >you can already do that by setting /etc/default/cpufrequtils from the > >cpufrequtils package. > > > >cheers > > > Yes, you can. However, I was arguing for a change in the defaults, > so that the average user would get a better performance with the > default settings. this bug then belongs to the package that sets ondemand as default governor, which isn't the kernel. cheers -- mattia :wq! -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
