[hylafax-users] New way of using Eicom diva boards for fax reception

["Luke McKee" <hojuruku@xxxxxxxxx>]

> I'll post again later regarding your apparent lack of knowledge about the
> Eicon boards and their proprietary API.  However, a couple of points in
> particular requires some education;

The lab results are not in yet?

Unless you test the CAPI interface of a eicon board and confirm
defects in fax reception in CAPI mode the criticism of that product
will have to be held back until the results are in. I don't really
care because I don't own one - I was just trying to help out other
hylafax users.

> Every evening our boards are soak tested in PC racks that support 16 PCI
> boards.  During OCTO+ testing we have 128 ports all sending and receiving
> fax's at the same time.  These machines are quite old running 1.2GHz PIII's
> with 512MB's of RAM.

The number of interrupts a PC support with IO-APIC is 224, sorry my
memory missed a digit in my last post. Even if you do perform
interrupt sharing, your still going to get heaps of them running that
many modems.

In any case under a high load in the example given there should be a
very high proportion of system interrupts. With 128 modems running
simultaneously on a Pentium 3/4, what would the system interrupt level
be? - and that's even before you count the software controlling the
serial port. The interrupt jitter and PCI bus contention could be a
problem too if you want something responsive, as a serial port locks
the whole pci bus to transmit 8 bits

http://tldp.org/HOWTO/Serial-HOWTO-3.php

I have one zaptel card. It's doing about 1000 interrupts a second,
which is about 0.7-1.2% of my CPU in system interrupts, just doing
polling and analysis of a DSP/DAA.

Now lets use the example with a modem port.
115200/8/16 = 900 number of interrupts per second per one port whilst
transmitting.
57600bps ~ 450 interrupts per second.
Assumption 16550A polls for interrupt every 16 bytes.
128 modems transmitting or receiving data at 57600 bps actually equals
57600 system interrupts per second!

(450*128 * 1%)/2 assuming the half the amount of processing is
required per modem port as a zaptel card uses in the kernel for
DSP/Fax/etc then my 1.8 AMD Ghz system needs to be 2-3 times faster to
cope with that load - and that's if the motherboard bus can cope with
the contention issues. Did I leave anything out in my appraisal of the
performance of 16550A serial architecture?

So I'm not disputing weather you could get that many modem ports
detected and operational in an idle state. But simultaneously? If you
don't use normal 16550A UARTS then maybe the above example doesn't
apply to your boards. What type of UART do your boards use exactly?
Are they enhanced in some way?

Does Sideband have anything to do with MSI-X on PCI express?

Cheers,

Luke

____________________ HylaFAX(tm) Users Mailing List _______________________
 To subscribe/unsubscribe, click http://lists.hylafax.org/cgi-bin/lsg2.cgi
On UNIX: mail -s unsubscribe hylafax-users-request@xxxxxxxxxxx < /dev/null
 *To learn about commercial HylaFAX(tm) support, mail sales@xxxxxxxxx*