On Wed, Jul 21, 2010 at 1:49 AM, Pieter Palmers <[email protected]> wrote: > I cannot help but noting that the 'most bone-headed way of doing low-latency > audio' called firewire provides timestamps related to the sample clock for > midi messages (even each MIDI byte). The timestamps are valid accross > multiple devices.
As the person that last uttered the words "bone headed" and "firewire" in the same sentence, I figure I should respond. :-) Yes, it's attempted in software, e.g. isosynchronous packets ... and it's full of fail: http://www.docstoc.com/docs/23936431/FFADO-firewire-audio-for-Linux-httpwwwffadoorg (Hmmm... I guess they do have interesting papers at the Linux Plumbers Conf.) "No direct access to devices hardware audio sampling clock (compare to PCI devices)" "Devices don't lock audio clock to iso clock" "Must infer audio clock relative to iso clock (and/or wall time) based on incoming time stamps" "must compensate for buffer and transport delays through firewire and device electronics" "some devices can be very picky about the accuracy of the inferred audio clock" "most manufacturers use vendor-specific extensions for key aspects of device control" "some manufacturers think they can do better than the standard and use their own" ............ http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.119.1433&rep=rep1&type=pdf --> complexity fail... jitter fail :: "The results of the test bench shows that RT- net over RT-FireWire gives performance compara- ble with RTnet over Ethernet. The former shows larger latency and larger latency variation (jitter), i.e. the difference between 97% threshold latency and worst case latency. This is due to the rela- tively more complex software stack: the packet path through the whole RT-FireWire includes more task handover, context switches, which inevitably cause more lantency to the whole data path. On the other hand, the former has a less leaning slope. This is due to the higher data transfer rate on FireWire: the FireWire devices in experiment can transfer data at 400Mb/s, while the Ethernet devcie in experiment can only transfer at 100Mb/s." ........................ -- Niels http://nielsmayer.com _______________________________________________ Linux-audio-dev mailing list [email protected] http://lists.linuxaudio.org/listinfo/linux-audio-dev
