This topic describes quality issues in a Cisco Unified Communications Manager multisite deployment.
IP networks are not designed to carry real-time traffic:
- Packet-by-packet delivery
- Packets can take different paths.
- There is no guarantee of the correct order.
- The problem is solved by RTP sequence numbers.
- Bandwidth is shared by multiple users and applications.
- Unpredictable available bandwidth.
- During peaks, packets need to be buffered in queues.
- Causes variable delays (jitter).
- Packets are dropped in case of buffer congestion.
- Congestion is likely on highly loaded links such as IP WAN, which is used between sites in a multisite environment.
- Jitter and packet drops impact voice quality.
IP networks are not designed to carry real-time traffic. Because of the nature of the network and packet-by-packet delivery in which each packet could take a different path, there is no guarantee that packets will arrive in the correct order at the destination. RTP sequence numbers resolve this issue.
Another issue is the fact that multiple users and applications share the bandwidth, and the actual required bandwidth varies significantly, even over short lapses of time. Therefore, the bandwidth that is available for Cisco Unified Communications Manager traffic is unpredictable. During peaks, packets need to be buffered in queues. If the congestion occurs for too long, buffers get filled up and packets are dropped. Higher queuing delays and packet drops are more likely on highly loaded, slow links, such as WAN links that are used between sites in a multisite environment. As a result, quality issues are common and need to be resolved by implementing QoS. Otherwise, voice and video packets are subject to variable delays (jitter) and packet drops, both of which impact voice quality.
Quality Issues Example: Jitter and Packet Drops
The figure illustrates how packets are queued during congestion.
During congestion, packets are buffered in queues. If a queue is full, packets are dropped.
During peaks, packets cannot be sent immediately because of interface congestion, so they must be stored in a buffer (“queued”). The time that the packet waits in such a queue is referred to as the queuing delay. The length of this delay can vary widely. If the queue is full, newly received packets cannot be buffered, so they are dropped. This action is called “tail drop.” Without any special treatment of voice packets, as in a FIFO processing model, the resulting jitter and packet loss decrease voice and video quality.