[linux-audio-dev] Davis votes for MP, news at 11

New Message Reply About this list Date view Thread view Subject view Author view Other groups

Subject: [linux-audio-dev] Davis votes for MP, news at 11
From: Paul Davis (pbd_AT_Op.Net)
Date: Thu May 17 2001 - 16:04:06 EEST


I had a chance to get out into the green of the local ridges of the
Appalachians yesterday, and was able to think a little about the LAAGA
issue away from my computer. After some consideration of the issues
raised here over the last few days, I feel moderately comfortable
with the idea that an inter-application glue API could use a
multiprocess model. Reasons I'm comfortable:

  * the overhead is low-ish
  * the overhead will decrease as processor speeds increase
  * it solves the GUI problem
  * we may all be running on Tera systems or 64 bit single address
      space OS's sooner than we realize

Reasons why I have doubts:
 
  * the overhead is not, under even the best circumstances, insignificant
  * the MP model is less stable in the face of machine load
  * the MP model will not encourage RT-style code design as strongly
      as the SP model

However, as Steve (and Abramo and Jim and ...) have pointed out, it
doesn't seem that hard to have a single body of "dsp" code that can be
used with either approach. This doesn't apply to the GUI problem,
however, and for me, this is perhaps the single most critical part of
my reason for liking the MP model.

I cannot imagine how I would write GUI's in the style I use (at the
source code level) if the GUI had to be in a separate process from the
actual "dsp" code. My use of libraries like libsigc++ to provide
anonymous callbacks makes MVC programming a joy, and this would
totally break down in the face of having to use IPC for communication
between Models, Views and Controllers. I have spent some time recently
wondering how I could write the GUI for ardour if it had to run in a
different process from the audioengine, and to be perfectly honest, it
seems like a nightmare. Now, some nightmares are worth implementing
anyway, but I think that the numbers from Abramo's test program
suggest that the cost of just going MP make this an acceptable way to
avoid the problems on this side of things.

Anyone suprised? :)

--p


New Message Reply About this list Date view Thread view Subject view Author view Other groups

This archive was generated by hypermail 2b28 : Thu May 17 2001 - 16:29:01 EEST