[clue] [tech] Asterisk in a virtual machine, timer, timing, & real time clock questions

[Jim Ockers]

Hi CLUEbies,

I want to run our office PBX/phone application server in a virtual 
machine.  The hardware machine running the PBX right now is big, noisy, 
ugly and a power hog, and we hate it even though it works fine.  I 
cloned the PBX (which is based on CentOS 4.3) into a virtual machine and 
fired it up.

Unfortunately I forgot to stabilize the system clock, and by the end of 
the day the phones were unusuable because the VM system clock/time was 
wandering around and even ntpd in the VM refused to work, insisting that 
it did not know what time it was.

Does anyone in CLUE-land have experience with asterisk PBX running in a 
virtual machine?  What are the do's and don'ts for this?  Do I really 
need a hardware clock or a Zaptel board?  I've read some papers like 
http://www.usenix.org/publications/login/2007-02/pdfs/chak.pdf which 
states "In  Asterisk, conferencing requires a reliable clock to mix 
audio. Such a timing  source is usually featured in the Digium hardware 
PCI cards, or one can
choose to use the ztdummy driver. In Linux 2.6, ztdummy defaults to 
using the kernel's real-time clock ..."

We have no Digium/Zaptel hardware in the server running the 
virtualization software (VMWare, in this case).  Even if I did, I 
wouldn't want to give the VM access to it, because having 
hardware-specific dependencies in a virtual machine kind of defeats one 
of the purposes of virtualization.

I've stabilized the system time in the VM by adding "clock=pit" to the 
kernel's boot command line and running ntpd in the VM.  The VM system 
time appears to not diverge from the host time in my unscientific 
observations (running "date" once in a while).  The ztdummy module is 
loaded.  Ntpd gets happy (stratum 3) within a couple of minutes after 
the VM boots up and stays that way indefinitely, note "stratum=3" below:

[root at pbx ~]# ntpq -c rv
assID=0 status=06a4 leap_none, sync_ntp, 10 events, event_peer/strat_chg,
version="ntpd 4.2.0a at 1.1190-r Mon Feb 21 17:54:52 GMT 2005 (1)"?,
processor="i686", system="Linux/2.6.9-34.ELsmp", leap=00, stratum=3,
precision=-19, rootdelay=124.990, rootdispersion=78.539, peer=2814,
refid=64.6.144.6,
reftime=d1d870bf.b879a283  Mon, Jul 25 2011 16:46:23.720, poll=10,
clock=0xd1d87161.073cb3e5, state=4, offset=0.569, frequency=-16.721,
noise=4.704, jitter=2.104, stability=12.984

[root at pbx ~]# ntpq -np
     remote           refid      st t when poll reach   delay   offset  
jitter
==============================================================================
 50.28.8.223     .STEP.          16 u    - 1024    0    0.000    0.000 
4000.00
 50.28.8.224     .STEP.          16 u    - 1024    0    0.000    0.000 
4000.00
*64.6.144.6      128.252.19.1     2 u  317 1024  377   98.730    0.569   
2.104

That offset is nice and small, a half second, which is close enough I think.

Since it's a business phone system, they don't want another disruption 
while the phones go off into the weeds for an afternoon to the point 
where we can no longer make & receive phone calls, and then have to 
revert back to the hardware PBX server anyway.

Is this [virtualizing Asterisk] even possible without /dev/zap or a 
hardware real time clock or timer hardware?  What do I need to do to 
ensure that this will work properly this time?  Any suggestions?

Thanks,
Jim

-- 
Jim Ockers, P.Eng. (ockers at ockers.net)
Contact info: http://www.ockers.net/msi.html


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://cluedenver.org/pipermail/clue/attachments/20110725/2be17838/attachment.html