7-3
Cisco Video Surveillance Solution Reference Network Design Guide
OL-29538-01
Chapter 7 Network Performance Considerations
Packet Loss
Packet Loss
Packet loss refers to the dropping of packets between a defined network ingress point and a defined
network egress point. Loss is detected by the reception of non-contiguous sequence numbers at the
receiver. Both TCP and RTP packets have a sequence number field in their respective packet headers for
this purpose.
In general, packet loss is caused by three main factors:
• Congestion – due to queue build-up and exhaustion of buffer space
• Lower-layer errors – bit errors, which might occur due to noise or attenuation in the transmission
medium
• Network element failure – link and device failures
When RTP data is transported over UDP, the sender is not notified of the packet loss because the
connection is on-way, sender to receiver, and there’s no concept of state. TCP, on the other hand, notifies
the sender through use of duplicate acknowledgements. The duplicate ACKs contain the sequence
number of the last contiguous packet received. If the lost packet did not make it to the receiver, the sender
discovers the packet loss when the retransmission timeout expires before the expected ACK is received.
Therefore, TCP is more reliable than UDP as a transport protocol; however, UDP is more efficient
because of lower protocol overhead. For high packet loss and high latency networks, TCP should not be
used as the transport protocol as it will only exacerbate a bad situation, further inhibiting real-time
delivery of data. Whenever congestion is detected, TCP slows the transmission rate to adapt to the
change and mitigate packet loss; however, when loss does occur, then the throughput is significantly
impacted as slow-start mode is invoked.
Note that since a single Ethernet frame (1500 bytes) can carry more than one IP video packet as payload,
the loss of a frame can have significant effects on the quality of the decoded stream, typically manifested
as pixelated video streams and gaps in recordings.
In order to effectively measure packet loss, the IP Video Surveillance network needs to be preconfigured
to monitor and report on the status of all media flows from video endpoints to media servers, and media
servers to client endpoints. This method can be characterized as the passive approach, in that
performance measurements are taken without disturbing the data operation, and are achieved through the
deployment of Cisco performance monitor.
Performance monitoring allows network administrators to detect video degradation due to packet loss,
before it significantly impacts the performance of VSM. Whenever a predefined threshold is crossed, a
user can be immediately notified either through a syslog message or SNMP traps, allowing for quick
fault isolation and resolution.
Mediatrace can also be used to measure packet loss along a network path, and on an on-demand basis.
When degradation in the stream quality is visually observed, or reported by the performance monitor,
the end-to-end path and the specific flow can be examined to determine which node along the network
is causing the loss. This is done by calculating metrics from values in the TCP, UDP and RTP headers at
each node. All nodes need to be configured as mediatrace responders.
More details on performance monitoring and mediatrace are discussed in the chapter on network
management.
Alternatively, packet loss along a network path can be measured on-demand through the use of synthetic
video traffic generated by IP SLA Video Operations probes. This is the active approach, since the IP SLA
VO probes emulate video endpoints by generating and sending realistic video traffic to receivers, along
the same network path that normal video traffic would take. As a result, the synthetic traffic is exposed
to the same path characteristics that real traffic would experience and therefore the packet loss metrics
collected are representative of the state of the network path