Technical data

DATA CENTER and CAMPUS NETWORKS DEPLOYMENT GUIDE
Deploying Brocade Networks with Microsoft Lync Server 2010 10 of 52
Packet Loss
Packet loss occurs when packets are sent but not received at the final destination, due to a network problem. Packet
loss is the proportion (in percentages) of packets lost en route across the end-to-end network. Packets can be
designated as lost for a variety of reasons: actual errors in transmission, corruption, packets discarded from
overflowing buffers or for having stayed too long in the buffer, and packets arriving with too much delay or too much
out-of-order to still be usable. However, the main reason for packet loss is discarded packets in congested routers,
either because the buffer is full and overflowing, or due to methods such as Random Early Detection (RED) or
Weighted Early Random Detection (WRED), which proactively drop packets to avoid congestion.
Well-sized and well-managed IP backbones and LANs are designed to operate at better than a 0.5 percent packet
loss average. Packet loss on end-to-end Internet routes, however, can occasionally reach 5 percent or even 10
percent. Wi-Fi connections can experience well in excess of 10 percent loss.
Several factors make packet loss requirements somewhat variable. Even with the same average packet loss, the
manner in which the packets are lost influences voice quality:
• There are two types of packet loss: random packet loss over time (single packets dropped every so
often during the call) and “bursty” packet loss (several contiguous packets lost in a short time window).
Losing 10 contiguous packets is worse than losing 10 packets evenly spaced over an hour.
• Packet loss may also be more noticeable for larger voice payloads (that is, packets representing a
longer time sample) than for smaller ones, because more voice is lost in a larger payload.
• Packet loss may be more tolerable for one codec over another, because some codecs have loss
concealment capabilities.
• Packet loss requirements are tighter for tones (other than Dual-Tone Multi-Frequency (DTMF) signaling)
than for voice. The ear is less able to detect packet loss during speech (variable pitch) than during a
tone (consistent pitch).
• Even small amounts of packet loss can greatly affect the ability of traditional TTY devices to work
properly, as well as transmission of faxes using the usual fax protocol T.30 over IP networks; standards
such as T.38 have been developed to reduce the impact of network issues on the reliability of faxing
over IP, but in practice they are not always supported, or the IP network may not be detected.
Jitter
Jitter is a measure of the time variability in the arrival of successive packets, generally expressed in milliseconds.
Jitter can result from packets taking different routes (for a variety of reasons, including load balancing or rerouting
due to congestion) and experiencing different propagation delays on those routes. Jitter can result from differences
in the effects of congestion, where some packets may have to wait for long buffer queues to be emptied, whereas
other packets may not. Jitter can also result in packets arriving out-of-order. Typically, the greater the network delay,
the greater the jitter, because each processing step is likely to add jitter.
The effects of jitter, if untreated, are similar to the effects of very severe packet loss at the endpoint, because
packets will arrive too late to be rendered to the end user. Therefore, the impact of jitter is reduced through the use
of a jitter buffer, located at the receiving end of the voice connection. The jitter buffer intentionally delays arriving
packets by more than the typical jitter value, in order to attempt to receive most jitter-affected packets, reorder
them, and retime them so that the end user hears the signal as intended. Unfortunately, jitter buffers introduce
incremental delay, which itself can negatively impact the experience. Therefore, jitter buffers typically contain only
about 20 to 40 ms of voice. Values of jitter in excess of the buffer length leads to packets being discarded.
By properly designing a network environment for peak loads, establishing proper QoS throughout the network, and
using Brocade Ethernet switches, you can reduce the amount of jitter. It is also important to properly provision the
network for adequate bandwidth so as to limit the amount of congestion, which can lead to increased jitter. When
deploying QoS, you should rate limit critical traffic so that other important data traffic can work in conjunction with