[VoiceOps] High Quality, Reliable Voice via the Internet / SIPNOC

[Christopher E. Brown]

Looking at it from the wrong viewpoint.

In this situation we are assuming that VOIP traffic is sharing a whole
series of transmit queues with no VOIP traffic, specifically general
Internet traffic.


VOIP and similar is relativly high PPS for the data rate, but is in
comparison to general data flows low volume (octets wise) and constant
(VBR action in codecs can cause a flow to vary in B/W by > 50% in short
time windows but the actual variance is small, generally single or
double digit kbps, not double or triple digit mbps.


If all the traffic in a queue was VOIP or equiv, than in the aggregate
of tens to thousands of calls the rate is very, very smooth and you can
go ahead and load that pipe to 95% or better on a 30 second average
without burst drop.


General data usage...  Very, very bursty.  These days a $350 laptop is
easily capable of generating 200KB bursts at 600 - 800 Mbit without even
trying.


Unless the core links handling bulk traffic are a large multiple of the
customer interface rates (real or emulated by a very granular shaper)
and on average have a good multiple of the largest customer service rate
free you are going to see burst drop

On 6/9/14, 6:00 PM, PE wrote:
> But isn't congestion caused by bytes and not number of packets? So, by
> that argument, larger packets will fill the queue faster than smaller
> and thus have a higher propensity to drop? And when it does, it is a
> bigger chunk of audio so it could actually reduce quality rather than
> improve it. 
> 
> Again, it's all theoretical unless someone has the tools, time, and
> motivation to publish an article with empirical data. 
> 
> 
> On Jun 9, 2014, at 9:45 PM, "Mark R Lindsey, ECG" <lindsey at e-c-group.com
> <mailto:lindsey at e-c-group.com>> wrote:
> 
> 
> 
> 
> On Mon, Jun 9, 2014 at 4:14 PM, Alex Balashov <abalashov at evaristesys.com
> <mailto:abalashov at evaristesys.com>> wrote:
> 
>     On 06/09/2014 02:50 PM, Mark R Lindsey wrote:
> 
>         2. Increase the ptime from 20 ms to 30-40 ms to reduce
>         packet-drop exposure
> 
> 
>     Or does this thesis lean on countervailing tendencies, such as
>     overall reduced PPS in a higher ptime scenario?
> 
> 
> You're on the right track with ptime. The theory idea is that:
> 
> (A) Most packet loss is due to congestion
> 
> (B) When congestion occurs the router selects a packet to drop
> 
> (C) The routers pick a packet to discard more-or-less at random 
> 
> (D) Therefore, A 180 byte packet is just as likely to be dropped as a
> 1500 byte packet. 
> 
> (E) A ptime=20 generates twice the packets as ptime=40, and therefore
> ptime=20 has twice the exposure to the discards 
> 
> (F) You can reduce your exposure to discards by reducing the number of
> packets you have in the queue.
> 
> (G) Reduced discards mean better audio quality.
> 
> _______________________________________________
> VoiceOps mailing list
> VoiceOps at voiceops.org <mailto:VoiceOps at voiceops.org>
> https://puck.nether.net/mailman/listinfo/voiceops
> 
> 
> _______________________________________________
> VoiceOps mailing list
> VoiceOps at voiceops.org
> https://puck.nether.net/mailman/listinfo/voiceops
> 


-- 
------------------------------------------------------------------------
Christopher E. Brown   <chris.brown at acsalaska.net>   desk (907) 550-8393
                                                     cell (907) 632-8492
IP Engineer - ACS
------------------------------------------------------------------------