Specifications

Performance Tuning, and Monitoring 159
As an example, assume a vocoder and rate of E-CELP at 9.6 kbps. The 9.6 kbps
represents the amount of data, minus headers, that will be used for the call
assuming no silence suppression. If frame packing is not used, the actual
bandwidth used is 13333 bps per call. So a PVC with 64 kbps CIR can only carry
four voice calls at the 9.6 kpbs rate.
Frame packing can be used to reduce the amount of bandwidth required for each
voice port - but be aware that frame packing introduces about 15 ms per
additional voice packet. Be aware that the maximum size packed voice frame can
be larger than the smallest fragment size allowed by FR. It wouldn’t be efficient to
pack the frames and then have FR fragment them so ensure that the packed
voice frame size doesn’t exceed the fragment size for the PVC.
5.1.3 PVCs Carrying Voice Only
The defaults for the voice adapter interface and FR interface should be sufficient
for a PVC which only carries voice traffic. A PVC carrying voice only does not
need to be fragmentation capable - it is only carrying voice and voice packets that
would not be fragmented anyway.
5.1.4 PVCs with Voice and Data
All PVCs that are not carrying voice but are on the same interface with a PVC that
is carrying voice must be enabled for fragmentation. The fragmentation size
depends on the access rate, the number of voice ports supported on the
interface, and the amount of delay that the user feels is tolerable in the voice
connection.
For example, with E-CELP at 9600 bps, the adapter will transmit a voice frame
every 15 milliseconds. It would be best to minimize the queuing delay for each
voice packet but some amount of queuing is tolerable. If you assume that a 30-ms
delay is tolerable, then the fragment size on the non-voice capable PVCs on a 64
kbps line should be around 240 bytes (64000 x 0.030 seconds). If the PVC is on a
T1 line, then the fragmentation may not be necessary if the 30-ms delay is
acceptable (1544000 x 0.030 = 5790 bytes).
5.1.5 PVCs Carrying Both Voice and Data
More tuning is needed when a PVC is carrying voice and data traffic. You will
need to tune the fragment size and the burst size to accommodate the voice/data
mix. Again, the access rate, the number of calls, and the vocoder and rate must
be considered.
Vocoder Rate 1 Packet 2 Packets 3 Packets 4 Packets 5 Packets
4800 bps 8533 bps 6666 bps 6044 bps 5733 bps 5547 bps
7470 bps 11203 bps 9336 bps 8714 bps 8403 bps 8217 bps
9600 bps 13333 bps 11466 bps 10844 bps 10533 bps 10437 bps
16000 bps 19733 bps 17866 bps 17244 bps 16933 bps 16747 bps