[asterisk-dev] Zaptel Interrupt Issues

[Bob]

I was reading all the #ifdefs incorrectly and assuming that the buffers in
zaptel were for the PPP and NET. It's pretty hairy in there.

I do have some Wanpipe machines for testing. The PCI hardware is mostly
irrelevant here. They all call zt_recieve and zt_tranmit 1000 times per
second. Oddly enough ... Wanpipe hardware generates an rx and a tx
interrupt, so it generates 2000 interrupts per second. Wanpipe also does a
TON of data movement in the usual places around the two zaptel calls. It
really affects their time in ISR very little.

In order to change the chunksize variable I had to use older cards for my
testing. These cards all have the old Tigerjet syle DMA in them. That means
that when I made the chunksize 512, all three of these drivers were moving
512 frames x 24 channels into the channel arrays in the interrupt context.
(512 * 24 * 2 (tx/rx) = 24576 data moves) This didn't come anywhere close to
doubling the time spent in the ISR. What I wound up with was a phone call
that sounded like RTP, but it was just to prove a point.

Whether I am covering up for other driver's kernel time, or just reducing
the context switching overhead seem irrelevant. Increasing the chunksize
drastically improves data integrity under heavy load. 

As previous threads on this topic have mentioned, there are much higher data
rate interfaces that don't have the same problem.

-----Original Message-----
From: asterisk-dev-bounces at lists.digium.com
[mailto:asterisk-dev-bounces at lists.digium.com] On Behalf Of Eric "ManxPower"
Wieling
Sent: Tuesday, October 02, 2007 1:08 PM
To: Asterisk Developers Mailing List
Subject: Re: [asterisk-dev] Zaptel Interrupt Issues

I'll bet his Zaptel has been patched by the Sangoma Wanpipe stuff.