On Feb 11, 2017 7:58 AM, "Jim Kusznir" <jim(a)palousetech.com> wrote:
Sorry for the delayed response, I finally found where gmail hid this
response... :(
So the application is FusionPBX, a FreeSwitch-based VoIP system, running on
a very unloaded (1% cpu load, 2-4 VMs running) system. I've been
experiencing intermittent call breakup, for which external support
immediately blamed on the virtualization solution claiming that "You can't
virtualize VoIP systems without causing voice breakup and other call
quality issues". Previously, I had attempted to run FreePBX
(asterisk-based) on a Hyper-V system, and I did find that to be the case;
moving over to very weak, but dedicated hardware, fixed the problem
immediately.
Since I sent this message, I did extensive testing with my system, and it
appears that the breakup is in fact network related. I've been able to do
phone to phone calls on the local network for extended durations without
issue, and even have phone to phone calls on external networks without
issue. However, calls going to my VoIP provider do break up, so it appears
to be the network route to my provider.
So, oVirt does not appear to be to blame (which I didn't think so, but was
hoping for some "expert information" to support this...It appears that I
got that and more with my tests).
Great to hear. I do believe that setting affinity and possibly taking into
account NUMA makes sense. Perhaps using SR-IOV is needed for low latency.
There is interesting work upstream qemu to improve throughout and reduce
latency in the expanse of more CPU usage.
Lastly, real time, mainly kernel and qemu-kvm, is also technology that
might be needed for some workloads. See [1].
Y.
[1]
https://mpolednik.github.io/2016/09/19/real-time-host-in-ovirt/
Thank you again for your work on such a great product!
--Jim
On Wed, Jan 4, 2017 at 10:08 AM, Chris Adams <cma(a)cmadams.net> wrote:
Once upon a time, Yaniv Dary <ydary(a)redhat.com> said:
> Can you please describe the application network requirements?
> Does it relay on low latency? Pass-through or SR-IOV could help with
> reducing that.
For VoIP, latency can be an issue, but the amount of latency from adding
VM networking overhead isn't a big deal (because other network latency
will have a larger impact). 10ms isn't really a problem for VoIP for
example.
The bigger network concern for VoIP is jitter; for that, the only
solution is to not over-provision hardware CPUs or total network
bandwidth.
--
Chris Adams <cma(a)cmadams.net>
_______________________________________________
Users mailing list
Users(a)ovirt.org
http://lists.ovirt.org/mailman/listinfo/users
_______________________________________________
Users mailing list
Users(a)ovirt.org
http://lists.ovirt.org/mailman/listinfo/users