150
Cisco Aironet 1520, 1130, 1240 Series Wireless Mesh Access Points, Design and Deployment Guide, Release 6.0
OL-20213-01
Troubleshooting
With the Cisco Mesh solution, the per-hop latency is less than 10 msecs, and the typical latency numbers
per hop range from 1~3 msecs. Overall jitter is also less than 3 msecs.
Throughput depends on the type of traffic being passed through the network. Traffic can be User
Datagram Protocol (UDP) or Transmission Control Protocol (TCP). UDP sends a packet over Ethernet
with a source and destination address and a UDP protocol header. It does not expect an acknowledgement
(ACK). There is no assurance that the packet is delivered at the application layer.
TCP is similar to UDP but it is a reliable packet delivery mechanism. There are packet acknowledgments
and a sliding window technique is used to allow the sender to transmit multiple packets before waiting
for an ACK. There is a maximum amount of data the client will transmit (called a TCP socket buffer
window) before it stops sending data. Sequence numbers are used to track packets sent and to ensure that
they arrive in the correct order. TCP uses cumulative ACKs and the receiver reports how much of the
current stream has been received. An ACK might cover any number of packets, up to the TCP window
size.
TCP uses slow start and multiplicative decrease to respond to network congestion or packet loss. When
a packet is lost, the TCP window will be cut in half and the back-off retransmission timer will be
increased exponentially. Wireless is subject to packet loss due to interference issues and TCP will react
to this packet loss. There is also a slow start recovery algorithm that is used to avoid swamping a
connection when recovering from packet loss. The natural effect of these algorithms in a lossy network
environment is to lessen the overall throughput of a traffic stream.
By default, the maximum segment size (MSS) of TCP is 1460 bytes, which results in a 1500-byte IP
datagram. Therefore, TCP fragments any data packet that is larger than 1460 bytes, which can cause at
least 30% throughput drop. In addition, the Cisco controller encapsulates IP datagrams in the 48-byte
CAPWAP tunnel header as illustrated in
. Therefore, any data packet that is longer than 1394
bytes is also fragmented by the controller, which results in up to 15% throughput decrease.
Figure 87
CAPWAP Tunneled Packets
CAPWAP
CAPWAP
CAPWAP
CAPWAP
CAPWAP
CAPWAP
CAPWAP
CAPWAP
CAPWAP
CAPWAP
205691