Technical data

DATA CENTER and CAMPUS NETWORKS DEPLOYMENT GUIDE
Deploying Brocade Networks with Microsoft Lync Server 2010 8 of 52
• Real-time metrics of the actual experience. Microsoft takes metrics to a new level and goes beyond
monitoring network metrics such as packet loss, jitter, and latency. Microsoft monitors the QoE of all
users on all calls by using Microsoft Lync Server 2010 Monitoring Server, which collects comprehensive
metrics and aggregates them in a Call Detail Record (CDR).
NETWORK PERFORMANCE CONCERNS
Users determine the ultimate measure of the performance of any service. In the case of voice, that ultimate measure
is the subjective, in-context perception of voice quality by the listener. Such subjective perception incorporates and
reflects intelligibility, clarity, pleasantness of speech, absence of defects, and overall conformity—as perceived by the
listener. It goes beyond simple restitution of the actual literal content to also include appropriate perception of
speaker identity, emotions, intonation, and timbre, as well as the absence of annoying effects. In addition,
perception can be affected by an almost limitless variety of effects (delay, background noise and interference,
clipping, distortion, echo, pops and clicks, and signal cuts or drops).
One of the key areas that impacts voice quality in a VoIP solution is the network infrastructure. This paper provides a
detailed performance profile and network metrics thresholds for Microsoft Lync Server 2010, to help network
integrators define and deploy network best practices that guarantee good media quality.
Voice Quality on IP Networks
Internet Protocol (IP) networks provide best-effort data delivery by default. Best effort allows the complexity to stay in
the end-hosts, so that the network can remain relatively simple and economical. The fundamental principle of IP
networks is to leave complexity at the edges and keep the network core simple. This approach scales well, as
evidenced by the ability of the Internet to support its host count and traffic growth without any significant change in
operation. If and when network services requests from hosts exceed network capacity, the network does not abruptly
deny service to some users, but instead degrades performance progressively for all users by delaying the delivery of
packets—or even by dropping some packets.
The resulting variability in packet delivery does not adversely affect typical Internet applications (bursty and
sometimes bandwidth-intensive but not very delay-sensitive applications such as e-mail, file transfer, and Web
“elastic” applications) until very severe network performance degradation. If data packets arrive within a reasonable
amount of time and in almost any order, both the application and the user are satisfied. Delayed packets are likely to
eventually arrive, because applications typically use Transmission Control Protocol (TCP) at the transport layer. Of
course, TCP is a connection-oriented protocol with built-in adaptation mechanisms to ensure error-free data transfer,
ordered data transfer, diagnostic, re-request, and retransmission of lost packets, discarding of duplicate packets,
and flow control (also known as congestion throttling). This makes TCP “unfriendly” to real-time applications such as
VoIP applications.
Real-Time Effective Bandwidth
The measure of the bandwidth of an end-to-end network path that is actually available to applications or network
flows at a given point in time is generally expressed in kilobits per second (kbps), On a shared network, this measure
fluctuates under the influence of flows generated by other applications, flows of the same application between other
users, or up- and downtime of network elements and links.
Delay (or Latency)
Delay is the measure of the time required for a voice signal to traverse the network. It is called one-way delay when
measured endpoint to endpoint. Round-trip delay, also called Round Trip Time (RTT), is measured end-to-end and
back. Delay is generally expressed in milliseconds. Delay results from the time it takes the system or network to
digitize, encrypt where appropriate, packetize, transmit, route, buffer (often several times), depacketize, recombine,
decrypt, and restitute a voice signal.
These sources of IP telephony delay can be grouped into four main categories: