[asterisk-dev] Re: Kernel modules => mainline kernel

[Steven Critchfield]

On Wed, 2007-02-14 at 17:15 -0500, Andrew Kohlsmith wrote:
> On Wednesday 14 February 2007 3:59 pm, Steven Critchfield wrote:
> > Seems to me that you are much more likely to get a more dense interface
> > card than a driver change to make it possible to put twice the normal
> > number of cards in a machine.
> 
> Not only that, but increasing the buffer length beyond that of a few frames 
> would allow us more leniency in the ISR.  One of the reasons the early TE 
> series cards had this high interrupt requirement was due to there being next 
> to no buffering; if we didn't get in there immediately and retrieve the data, 
> it'd be overrun.  Bigger buffers make for higher latencies, this is true, but 
> as another poster mentioned; if you're already hitting an ethernet 
> connection, you're not overly concerned about this.

Problem with a larger buffer is when you go from TDM to TDM. If you miss
that pickup, and now you are behind in your sends to the other endpoint,
did you send an empty frame for the timespot that was missed, or are you
further buffering to alliviate that potential miss?

Granted most of my installs are TDM to recordings, so a buffer doesn't
affect me. In fact, it would probably mean lower system requirements as
the buffer would cover some otherwise missed frames.
-- 
Steven Critchfield <critch at basesys.com>