#322: I/O modes
------------------------+---------------------------------------------------
Reporter: mrpingouin | Owner: admin
Type: Bugs | Status: new
Priority: 1 | Milestone: 1.0 beta
Component: Liquidsoap | Version:
Keywords: | Mac: 1
Linux: 1 | Netbsd: 1
Other: 1 | Freebsd: 1
------------------------+---------------------------------------------------
Description changed by mrpingouin:
Old description:
> I was reading the history of our Alsa classes, and it seems that we could
> clean traces of past attempts.
>
> It seems to me that IoRing is very good, it provides tight I/O. It
> clearly is better than the old Alsa code with active waiting and delays.
> Now that it supports a size=1 ring, it might also be as good as the
> unbuffered Alsa operators. Anybody using that one?
>
> The old IoRing buffer is still used for buffered I/O in opal. This is
> done in the old style, active waiting and delays. Couldn't we use IoRing
> there? Or is Opal dead anyways?
>
> While this kind of cleanup makes sense before the next release, it's also
> OK to wait, especially since we might be working more on
> synchronization/buffers/clocks after the release. For example, in my view
> we should actually only keep the unbuffered Alsa in the future, and
> create a pure IoRing operator that we would combine. It's kind of scary
> to touch that again; I hope we'll not loose any of the progress we've
> painfully made so far.
New description:
I was reading the history of our Alsa classes, and it seems that we could
clean traces of past attempts.
It seems to me that IoRing is very good, it provides tight I/O. It clearly
is better than the old buffering code with active waiting and delays
(defeating the purpose of buffering by interfering with the lock). Now
that it supports a size=1 ring, it might also be as good as the unbuffered
Alsa operators. Anybody using that one?
The old Ringbuffer module is still used for buffered I/O in opal. This is
done in the old style, active waiting and delays. Couldn't we use IoRing
there? Or is Opal dead anyways?
Now that we have clocks, we could also only keep the unbuffered Alsa (and
Opal, etc) in the future, and create a pure IoRing operator that we would
combine.
It's kind of scary to touch that again; I hope we'll not loose any of the
progress we've painfully made so far. Making some careful measures of time
drifts in various situations should be helpful.
--
--
Ticket URL: <http://savonet.rastageeks.org/ticket/322#comment:4>
Savonet <http://savonet.rastageeks.org/>
Let's program our stream !
------------------------------------------------------------------------------
_______________________________________________
Savonet-trac mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/savonet-trac
