On 7/14/15 11:04 AM, Alexander E. Patrakov wrote:
14.07.2015 20:03, Mark Brown wrote:
Like previous years, we're going to hold a meeting to discuss lowlevel audio on Linux. This will be held the day after ELC Europe on 8th October at CCD (The Convention Centre) in Dublin.
If you're interested please sign up in the attendee list:
https://docs.google.com/spreadsheets/d/1bi-8GHsqlzt41FR20WiMuWILe_cZowFireVM...
and please follow up to this mail with any topics you'd like to see raised so we can start collecting them (probably another Google doc will be forthcoming for them).
I would like to see some direct measurements related to recent power-saving proposals, including the "disable rewinds" flag. Myself, I can redo battery-life measurements on Intel-based laptops that my colleagues have, and maybe compare dmix, PulseAudio and CRAS in terms of power consumption.
Ideally, I would like to reevaluate the design decisions (namely, the need to keep the system responsive to new streams while keeping the average wakeup rate as low as possible, IMHO even to the point of "absurdly low") that led to the need to support rewinds (and the associated complexity) in the first place. Reminder: on my Sony VAIO VPC-Z23A4R laptop and hw:0 device, in an unrealistic test with the screen turned off, wi-fi turned off, and the SSDs put to sleep, going from 200 to 1 wakeup per second saved only 935 seconds of battery life out of 25742.
I had more success that you in my experiments a long time ago, but this really depends on what the rest of the system does. You have a point though that we should talk about design decisions for userspace code. The current PulseAudio implementation lags behind the capabilities of newer devices with two or more outputs (low-latency, increased buffering and hardware/firmware mixing). This should really be a topic following the presentation of the HDAudio/Asoc restructuring that Vinod mentioned (i.e. how to change userspace code to support driver capabilities). -Pierre