-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On Sun, 13 May 2007 19:58:50 +0100,
Mark Kent <mark.kent@xxxxxxxxxxx> wrote:
> Jim Richardson <warlock@xxxxxxxxxx> espoused:
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>>
>> On Sat, 12 May 2007 17:57:09 +0100,
>> Mark Kent <mark.kent@xxxxxxxxxxx> wrote:
>>> [H]omer <spam@xxxxxxx> espoused:
>>>> Verily I say unto thee, that Roy Schestowitz spake thusly:
>>>>> [X-Fi Drivers to Linux]
>>>>>
>>>>> ,----[ Quote ]
>>>>> | Creative plans to make proprietary (closed source) drivers available
>>>>> | for the X-Fi series of sound cards. SEE Soundcard Support for
>>>>> | updated details. These drivers should have full support for ALSA
>>>>> | (playback, recording, mixer, MIDI, synthesis) and OpenAL 1.1 (with
>>>>> | EAX effects).
>>>>> `----
>>>>>
>>>>> http://opensource.creative.com/
>>>>
>>>> Creative still haven't got their *Vista* drivers for X-Fi out of Beta,
>>>> and they're already going ahead with Linux drivers!
>>>>
>>>> My, my. How times change.
>>>>
>>>>> Linux is suitable for audio professionals.
>>>>
>>>> With a low-latency kernel, better ... even.
>>>>
>>>
>>> I saw some excellent work on kernel patches done by one of the main
>>> embedded linux providers (don't recall which one); the blocking
>>> behaviour of the kernel was hugely improved.
>>>
>>> The problem of streaming audio through the kernel is much the same as
>>> the problem of moving streams across networks - you need a particular
>>> kind of system behaviour which is able to fully respect the temporal
>>> relationship of the data constantly, not just in bursts. We need 3-mode
>>> kernels...
>>>
>>
>>
>> What's a 3 mode kernel? user, kernel and other?
>>
>
> File, stream and message. Streams need to be handled differently from
> files and messages, since the temporal relationship of the information
> units is critical in streams, as the brain cannot cope with jitter,
> packet loss or wander beyond anything but a tiny amount.
>
> This is much worse in conversation, where beyond about 250ms of end to
> end fixed latency also makes conversation very difficult, and beyond
> about 400ms it becomes almost impossible. Discernable echo beyond 25ms
> delay makes even speaking virtually impossible for most people.
>
> Messages, such as telnet/ssh, or file transfers like http, nntp, ftp,
> scp, nfs and so on do not care very much about the temporal relationship
> being maintained consistently - it can be mildly irritating to await a
> prompt or a listing, or see file transfers slow to a crawl for a while,
> but it doesn't make usage near impossible.
>
> The net result is that connectionless packet switching cannot be used
> for streams in networks, since deterministic performance cannot be
> guaranteed - a lot of people think that if you over-dimension, it'll
> work, but it doesn't and cannot. The same capability is required in the
> kernel, ie., deterministic behaviour. It's not about speed or bandwidth,
> but about predictability.
>
> Look up the PBB-TE article on wiki for more, if you're really interested.
>
Sounds interesting, I will. Thanks for the info.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
iD8DBQFGR2URd90bcYOAWPYRAl2nAJ9UGAZjtVoM6ZPkRss0oErXc/zVkgCglMdu
X0ytLjFrx4uiORcE/jK2fAw=
=VU2B
-----END PGP SIGNATURE-----
--
Jim Richardson http://www.eskimo.com/~warlock
I have plenty of talent and vision. I just don't give a damn.
|
|
