Skip to contentSkip to conversation

buffer_size and period_size

DISCUSSION Mark Knecht 1 participant 1 message Jan 27, 2003 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw
Quoted 8 lines

-----Original Message-----
From: Paul Davis [mailto:[email protected]]
Sent: Monday, January 27, 2003 10:39 AM
To: Mark Knecht
Cc: [email protected]
Subject: Re: [Alsa-devel] buffer_size and period_size

Quoted 8 lines
Quoted 3 lines

Paul,
So with -p 64 -n 2 settings, what number of bytes of audio data is
transferred across the PCI bus between each interrupt?

the period size is always (as with almost all ALSA metrics) in units of audio frames (1 sample for each channel). so with JACK running at -p 64, 64 frames of data are transferred. how many bytes that is depends on the sample width and the number of channels.

Cool, so this says my RME HSDP puts out 26 samples 64 times, and then signals with an interrupt, while my Hammerfall Light puts out 18 samples 64 times? If true, then both cards interrupt at the same rate, but the HDSP 9652 uses more PCI bandwidth. (26/18)

Presumably each sample uses 24-bits of audio data, packed in a 32-bit hardware DWORD moving across the PCI bus? This would be clean and easy for software to deal with. If true, then the number of bytes would be 4 bytes/channel * 26 channels * 64 frames = 6656 bytes across the PCI bus between interrupts. (Or desire to interrupt, not necessarily interrupt services.)

Quoted 9 lines
Quoted 4 lines

Also, does the number of bytes transferred change based on how many
channels are enabled? Or does my RME always transfer 26 channels of data even if I am not using some channels?

its hardware dependent. for the RME, the same amount is always moved, regardless of how many channels are enabled. in the case of the RME Hammerfall series, its more accurate to say that you cannot change the number of channels enabled. this is not true of all other hardware.

Sure. Sensible. They just always send 26 channels of data. The loading on the PCI bus is therefore independent of how many channels I'm actually using.

Quoted 9 lines
Quoted 3 lines

I am assuming that a card like the RME is a bus master, moves so many
bytes, and then interrupts to tell the system that the bytes are

there. Is

Quoted 1 line

this basically the case?

yes. keep in mind that this only covers 1 direction: the interrupt also tells the system that its possible to put data into the "hardware buffer".

Good point. Thanks!

-------------------------------------------------------
This SF.NET email is sponsored by:
SourceForge Enterprise Edition + IBM + LinuxWorld = Something 2 See!
http://www.vasoftware.com