Re: [hylafax-users] Multi serial solutions

[Andrew Rinaldi <andrew.rinaldi@xxxxxxxxxxxx>]

Michael Hallager wrote:
> > Andrew Rinaldi wrote:
> >     
> > > choose a board that has a "PCI Status Register"...
> > >       
> > Sorry, that should have said "PCI Interrupt Status Register".  This
> > register allows the serial driver to examine the status of all board
> > interrupts without polling each port in tern.  This register was a
> > standard feature of the first multiport serial cards when they were
> > launched in the 1980's and is invaluable for optimising interrupt
> > service routine performance.
> >
> > This is one reason why we can run 128 fax ports simultaneously on a
> > 1.2GHz PIII machine.
> >     
>
> Hello Andrew-
>
> I only want to run 4 or 8 ports. Non of the cards I have looked at have this 
> feature. Is it important at this port-scale?
>
> Also - you say "board interrupts" - the board I'm looking at only uses 1 IRQ.
>
> Kind regards,
>
> Michael.

Hello Michael,

If your multiport serial board does not support a 'interrupt status 
register' then the multitude of interrupts that are generated by each 
and every port are all routed to a single IRQ.  This simplistic 
architecture forces the Interrupt Service Routine (ISR) to poll each 
port on a multiport board in turn looking for the source of any 
interrupt that it receives.  This interrupt polling is extremely 
inefficient and therefore historically multiport serial boards included 
a 'interrupt status register' to optimise the performance of the ISR by 
allowing it to read the status of all the ports on the board with 
a single read.  Unfortunately, the 'interrupt status register' has been 
removed from most modern multiport serial boards.

To some extent the Linux serial driver is also a 'moving target' with 
some significant new features (and bugs) being introduced during the 
last few years.  Here are a couple of issues that you may want to 
consider when deliberating over your proposed multiport solution;

(1) hardware flow control

The serial driver has a number of problems whereby the hardware flow 
control on the serial interface fails during heavy workload.  If I 
remember the details correctly, this bug is most prevalent during IDE 
disk interrupts when heavy modem flow control is in use;

http://lkml.org/lkml/2007/7/26/46

(2) The Interrupt Service Routine (ISR) seems to have a 'race' condition

Some of the reports of modems 'wedging' are, we believe, due to a race 
condition in the ISR.  Unfortunately, this 'wedging' is so dependant on 
Kernel version, motherboard, chipset, and modem type that we have never 
been able to characterise the bug.  The workaround that we have found is 
to use the 'interrupt status register' to read multiple interrupts 
from the board in a single read.... we suspect that this simply reduces 
the workload on the ISR and makes the race condition so rare that it can 
be ignored.


Needless to say, Mainpine implement multiple workarounds for these (and 
other) issues.... it might be that our software port reset, interrupt 
status register, and firmware watchdogs, will help you deliver a robust 
and reliable fax server.

Regards

Andrew Rinaldi
Mainpine Developer Support
USA +1 866 363 6680 | UK +44 1225 807 807
andrew.rinaldi@xxxxxxxxxxxx | www.mainpine.com


____________________ 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*