Increasing payload size of voice packet?

[Leonlei]

The typical payload size of voice packet in G.729 codec is 20Bytes and the total bandwidth (including 5% overhead) of each call is 32.67Kbps. It lookd like increasing payload size can reduce the required bandwidth for each call based on the result of running Cisco Codec Calculator. As far as I know one drawback of increasing payload size is to add more process time on router and cause some more delay. Does anyone have practical experiencing with changing the payload size?

[joe19366]

You dont want to do it!

the reason its so small (yes small means more overhead) is to avoid serialization, packetization and queue delays.

By making voice larger than the default small packet size of 20 bytes, you are increasing the amount of speech sampled and transmitted in one packet.

Even if you are using llq or other voice friendly queuing strategies (pq, etc) you run the risk of a single packet's loss exceeding the capabilities of the receiving device's de-jitter buffer.

Also, to avoid jitter from serialization delay with frame-relay frf.12, PPP multilink LFI, our goal is to limit packet size to equal to one tc interval. We NEVER want voice packets sitting in a software queue for too long behind a larger data packet waiting to get onto the tx-ring (hw-queue). By increasing the voice packet size beyond what very well may be one tc interval on slower speed links, you are subjecting voice to fragmentation interleave operations on say a mlppp bundle. This could cause delays on the receiving end, leading to jitter!

I would read the Wendell Odom QOS exam certification guide for some more clarity on these issues.

Thanks,

Joe