[Asterisk-Users] PCI front mount chassis?

[Bruno Haas]

Wow, 1000 IRQs a second. I'm quite amazed. Does anybody know which 
applications would require such a low latency ? It does seem to me that 
this way of doing things is rather dangerous and prone to problems. 
Anybody can comment ?

Thanks
Bruno

Nicolas Bougues wrote:

>On Fri, Mar 12, 2004 at 08:41:36AM -0500, Walt Reed wrote:
>  
>
>>The voice cards generate an order of magnitude more interrupts than
>>anything else. This "may" be why it's not recommended to share
>>interrupts on voice cards. Don't know if the T1 cards have a similar
>>issue. I would hope not. The x100p's are a pretty simplistic device.
>>They probably generate an interrupt for every byte. The x100p's are also
>>used for timing in things like MOH and MM conferences AFAIK. It seems
>>like it would be nice to only put one card in "timer mode" if that is
>>indeed what is generating all those interrupts. Could someone "in the
>>know" enlighten us?
>>
>>    
>>
>
>Digium boards usually generate 1000 interrupts/sec. This is mostly a
>"timer" interrupt, so that the driver can poll the board for 8 bytes
>per channel 1000 times per second. The polling is either slave (PIO),
>or busmaster (DMA).
>
>Each board generates 1000 interrupts/second, no matter the kind/number
>of ports on the boards. The driver knows if that's a single channel
>board, with 8 bytes to fetch, or quad E1, with 128*8 bytes to get on
>each interrupt.
>
>There are quite strict timing requirements : if one interrupt is lost,
>1/1000th of sound on the line(s) is lost, which can be quite bad for
>things like HDLC (on T1/E1) or modem sound (no matter the kind of
>channel).
>
>Sharing an IRQ may introduce some latency, because the IRQ has to be
>handled by several drivers, and if each of them does not behave
>properly, frames can be late/lost.
>
>  
>