Re: flexfax: Volume questions

[David Birnbaum <davidb@chelsea.net>]

On Fri, 4 Sep 1998, Nico Kadel-Garcia wrote:

> > We're getting ready to deploy a rather large fax application...I was
> > wondering, does anyone have any practical experience with the Hylafax
> > software and how much it can handle?  We're looking at a four-CPU SPARC
> > machine running Solaris, with potentially 96 modems attached to it, and
> > potentially several of these machines.  We're looking to use Hylafax,
> > since it gives us the source to put in our own hooks and enhancements, but
> > I haven't heard of anyone using it with more than 8 modems attached to it.
> 
> Bless you! What a wonderful use of freeware! Have you looked into what
> brand of modem you will use?

We've had good results with the eight USR modems we have attached to it
now, but I'll be damned if USR gets another $ of mine the way they've
treated us up until this point.  I have a Hayes Century chassis (Rockwell
based) lying around with 16 modems that we'll expand with temporariliy,
but when we get serious I'll probably price out high-density rack
solutions that seem to have the best fax firmware.

> > Any comments before we get started?  As a side effect of doing this with
> > Hylafax, we'll probably have some nifty features of our own to release
> > back into the Hylafax community, but it it would be nice to hear of others
> > treading some of this ground in the past.
> 
> Are you supporting dial-up modem use, or fax service only? I assume
> that you are running Solaris 2.6: be aware that the asppp
> implementation under Solaris was famous for difficulties in previous
> versions of Solaris.  You may want to proceed directly to ppp-2.3.5,
> at ftp://cs.anu.edu.au/pub/software/ppp/.

No, purely fax in/out.  We have a ton of other equipment (Ascend, cisco) 
that handles the PPP dialup much better than Solaris ever could.

My main concerns with scale is to find out if there's any coded limits
within Hylafax in terms of structures that limit the numbers of modems
attached.  I've also noticed that the faxq process seems to take more CPU
time than it should, considering the scheduling that it's doing.  The
script auxillary stuff (notify, sendfax, and the like) are all being
replaced with higher-volume things, but I don't want to muck too much with
the engine unless we have to.  I have noticed some flakiness with the job
management within Hylafax that we're going to look into, and we'll
probably change the jobID from an int to a longit (with a 1M+ duty cycle
monthly, an int don't cut it!)  As it is now, just doing about 4K a day,
it's already getting to be a pain in the rear to track things.  We're also
going to have to do some cleanup of the docq, doneq, log, and info
directories to break them down (every try to do an ls of a directory with
100K files in it?  Bad idea....) 

Anyway, as this progresses over the next few months, I'll probably be
posting a lot more notes to the list ;), not to mention the functional
PERL OO interface, which is about 60% finished at this point.  Hopefully
we can see some cool new functionality added to the Hylafax package as a
result of this whole thing.

Cheers,

David.