1. Performance of VoIP in a 802.11b wireless mesh network
D. Niculescu, S. Ganguly, K. Kim, R. Izmailov
NEC Laboratories America

2. 802.11 wireless mesh natural extension of one-hop wireless APs
useful in home and enterprise
advantages:
reliability/recovery
scalability
load balancing
D. Niculescu
Performance of VoIP in

3. Multihop disadvantages
complex implementation
requires route maintenance
potential performance degradation
traffic from the same flow interferes with itself
D. Niculescu
Performance of VoIP in

4. Testbed Nodes Intel Stargate (Crossbow) linux 2.4
802.11b wireless cards x 2
15 nodes in 30x50m building
Voice traffic
Zyxel Prestige VoIP phones
packet generators

5. Voice quality measure
R-score(loss, delay) is a codec specific quality measure
The wireless mesh can trade one for the other:
Loss can be reduced by retransmitting packets
Delay can be reduced by dropping packets
Example: G.729a
20 packets / second
20 bytes
max R-score 82%
loss < 2%
delay < 150ms
D. Niculescu
Performance of VoIP in

6. Challenges of maintaining voice quality in multihop
route changes
reduced capacity because of
self interference
high overhead for small voice packets
loss < 2%
delay < 150ms
D. Niculescu
Performance of VoIP in

7. How to improve QoS & capacity?
Three techniques
multiple cards – scalability, self interference
label switching – slow routing updates
aggregation/compression – overhead
D. Niculescu
Performance of VoIP in

8. Multiple cards Single channel Case A Case A Case B Case B 1 7 1 7 1
Case A
Case A
Case B
Case B

9. Voice routing: problems
used DSDV and DSR
loss based metrics
hop count
call across entire building (one card)
problems
temporary long links
slow rerouting
how reactive?
Conclusion:
standard routing is not appropriate for real-time traffic

10. Voice routing: label switching
collect most popular DSDV paths overnight
maintain best 5 paths for each S-D pair
monitor each path with a low rate ping
continuously switch to the best path
no packets are lost because of the switch
D. Niculescu
Performance of VoIP in

11. Voice overhead: 802.11, IP, UDP, RTP
G.711
G.729
D. Niculescu
Performance of VoIP in

12. aggregation & header compression
aggregation only:
several IP packets become payload for a large IP packet
simulation vs. testbed
aggregation + header compression: simulation only
header compression (cRTP style) provides little benefit
3x aggregation benefit; 2x header compression benefit
D. Niculescu
Performance of VoIP in

13. opportunistic aggregation
controlled delay is introduced at the ingress of a flow
only packets queued for next hop are aggregated
intermediate nodes do not delay to improve aggregation
short flows do not delay long flows
example
A, B calls aggregated independently
aggregation A+B should not delay A
network time
for flow A
A alone
61ms
A,B agg
89ms
A+B agg
79ms

14. implementation Click configuration Intel Stargate (PDA sized)
linux / 64M / 400Mhz
click modular router (MIT)
DSDV, DSR
aggregation modules
labels stored in IP header
packets can use default routing as well
current work: transparent mesh
D. Niculescu
Performance of VoIP in

15. summary experimental investigation of VoIP quality in mesh
three techniques explored
use of multiple interfaces
easiest scalability
challenges: available channels, allocation
label switching
seems best solution for real-time traffic
hard to give guarantees with unlicensed spectrum
aggregation && header compression
work much better together
13x capacity increase with all optimizations
D. Niculescu
Performance of VoIP in