Re: [hylafax-users] problems with faxq

[Lee Howard <faxguy@xxxxxxxxxxxxxxxx>]

As we don't yet know the specifics of this particular condition which 
Chris is reporting, and as there are a variety of ways in which the 
symptoms of which he speaks could arise, I'm unable to say with any 
certainty that the condition would or wouldn't occur for him with 6.0.x.

That said, with as many modems and jobs as Chris is discussing, it's a 
fairly trivial matter to get faxq in 6.0.x (including 6.0.5) to strangle 
the CPU resources on a well-endowed server.  I've done it on multiple 
occasions before (in the lab as I only use HylaFAX+ in production), but 
it's certainly not worth $5 to reproduce for you now.  We've been 
through this many times before.  I see no need to rehash it again now.

For what it may be worth I just ran a test with 100 iaxmodems (50 
sending to 50 receiving) on a HylaFAX+ 5.4.3 system with default 
configurations.  I submitted three 1-page documents (coverpage+2) to 
2000 unique destinations using 'sendfax -z list file1 file2'.  At no 
point has 'top' displayed CPU usage at or even near 99%.  Here's what 
the top few lines from 'top' look like when I see faxq running at its 
*highest* CPU usage:

---------------------------------
top - 15:34:57 up  6:40,  3 users,  load average: 11.42, 8.54, 4.19
Tasks: 363 total,   3 running, 349 sleeping,   0 stopped,  11 zombie
Cpu(s): 49.2%us, 42.4%sy,  0.0%ni,  5.2%id,  0.0%wa,  0.0%hi,  3.2%si,  
0.0%st
Mem:   3979908k total,   868200k used,  3111708k free,    30448k buffers
Swap:  2047996k total,        0k used,  2047996k free,   512068k cached

 PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  
COMMAND                                                                                                      

10355 uucp      20   0 11760 2620 1368 R 69.3  0.1   1:22.91 
faxq                                                                                                         

9469 root      20   0 48544  18m 3628 S 13.0  0.5   0:07.65 
asterisk                                                                                                     

15593 root      20   0  2908 1316  860 R  2.5  0.0   0:06.52 
top                                                                                                          

9579 uucp      20   0  3420 1012  836 S  1.9  0.0   0:02.11 
iaxmodem                                                                                                     

9518 uucp      20   0  3420 1012  836 S  1.6  0.0   0:02.12 
iaxmodem                                                                                                     

---------------------------------

You can see that the system load is quite heavy.  (It's actually too 
heavy to be healthy for timing-sensitive Class 1 software modems, and so 
this test is a contrived and liberally harsh scenario that should not 
come up in thoughtful production use.)  However, the system load isn't 
caused by faxq; the load is caused by 100 iaxmodems, 100 faxgettys, and 
Ghostscript running near-constantly.  When the fax starts and stops 
begin to become more-staggered over-time then the timings of the various 
operations in iaxmodem, faxgetty, and Ghostscript will not be so 
piled-up on-top of each other... then the system load will start to drop 
down until it reaches about 2.00, even while the queue of 2000 is still 
going with all 100 modems busy.

So while I do not doubt that Chris is reporting a bona-fide problem, I 
want to be clear that this kind of usage is tested in HylaFAX+, and so 
whatever is going on in Chris's scenario it is something that needs to 
be researched before jumping to conclusions about how it's happening.

Thanks,

Lee.



Darren Nickerson wrote:
> You probably won't be able to try this because you're in production, but if you can reproduce this 99% faxq condition on hylafax-6.0.x I'll send you $5 ;-)
>
>  (we rewrote faxq several years ago to address exactly this kind of problem).
>
> -d
>
>
> On Oct 21, 2010, at 6:41 PM, hchris@xxxxxxxxxxx wrote:
>
>   
> > Hi,
> >
> > I have running Hylafax+ 5.4.2 with Asterisk and 60 iax-modems on a Linux server
> >
> > If I add 2 or 3 thousand equal faxes with parameter -z faxnumberlist to the queue I got from top:
> > The faxq needs 99.7% to 100,2% of CPU
> > load is 3 to 5
> >
> > Is this normal? 
> >
> > Is there any possibility to run Hylafax+ with multiprocessing?
> >
> > Christian
> >
> >
> >     
>
>
> ____________________ 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*
>
>
>   


____________________ 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*