Telephone Network (PSTN). As an engineering and design issue, the placement of these nodes in the network becomes crucial. We will tackle this issue in design step 5. Other hardware requirements include a VoIP client terminal, which can be a separate VoIP device, i.e. IP phones, or a typical PC or workstation that is VoIP-enabled. A VoIP-enabled workstation runs VoIP software such as IP Soft Phones .
Fig. 3 identifies the end-to-end VoIP components from sender to receiver [9]. The first component is the encoder which periodically samples the original voice signal and assigns a fixed number of bits to each sample, creating a constant bit rate stream. The traditional sample-based encoder G.711 uses Pulse Code Modulation (PCM) to generate 8-bit samples every 0.125 ms, leading to a data rate of 64 kbps . The packetizer follows the encoder and encapsulates a certain number of speech samples into packets and adds the RTP, UDP, IP, and Ethernet headers. The voice packets travel through the data network. An important component at the receiving end, is the playback buffer whose purpose is to absorb variations or jitter in delay and provide a smooth playout. Then packets are delivered to the depacketizer and eventually to the decoder which reconstructs the original voice signal. We will follow the widely adopted recommendations of H.323, G.711, and G.714 standards for VoIP QoS requirements.
VI
Table 1 compares some commonly used ITU-T standard codecs and the amount of one-way delay that they impose. To account for upper limits and to meet desirable quality requirement according to ITU recommendation P.800, we will adopt G.711u codec standards for the required delay and bandwidth. G.711u yields around 4.4 MOS rating. MOS, Mean Opinion Score, is a commonly used VoIP performance metric given in a scale of 1–5, with 5 is the best. However, with little compromise to quality, it is possible to implement different ITU-T codecs that yield much less required bandwidth per call and relatively a bit higher, but acceptable, end-to-end delay. This can be accomplished by applying compression, silence suppression, packet loss concealment, queue management techniques, and encapsulating more than one voice packet into a single Ethernet frame.
3.1.1. End-to-end delay for a single voice packet
Fig. 3 illustrates the sources of delay for a typical voice packet. The end-to-end delay is sometimes referred to by M2E or Mouth-to-Ear delay. G.714 imposes a maximum total one-way packet delay of 150 ms end-to-end for VoIP applications . In [22], a delay of up to 200 ms was considered to be acceptable. We can break this delay down into at least three different contributing components, which are as follows (i) encoding, compression, and packetization delay at the sender (ii) propagation, transmission and queuing delay in the network and (iii) buffering, decompression, depacketization, decoding, and playback delay at the receiver.
3.1.2. Bandwidth for a single call
VII
The required bandwidth for a single call, one direction, is 64 kbps. G.711 codec samples 20 ms of voice per packet. Therefore, 50 such packets need to be transmitted per second. Each packet contains 160 voice samples in order to give 8000 samples per second. Each packet is sent in one Ethernet frame. With every packet of size 160 bytes, headers of additional protocol layers are added. These headers include RTP+UDP+IP+Ethernet with preamble of sizes 12+8+20+26, respectively. Therefore, a total of 226 bytes, or 1808 bits, needs to be transmitted 50 times per second, or 90.4 kbps, in one direction. For both directions, the required bandwidth for a single call is 100 pps or 180.8 kbps assuming a symmetric flow. 3.1.3. Other assumptions
Throughout our analysis and work, we assume voice calls are symmetric and no voice conferencing is implemented. We also ignore the signaling traffic generated by the gatekeeper. We base our analysis and design on the worst-case scenario for VoIP call traffic. The signaling traffic involving the gatekeeper is mostly generated prior to the establishment of the voice call and when the call is finished. This traffic is relatively small compared to the actual voice call traffic. In general, the gatekeeper generates no or very limited signaling traffic throughout the duration of the VoIP call for an already established on-going call. In this paper, we will implement no QoS mechanisms that can enhance the quality of packet delivery in IP networks. A myriad of QoS standards are available and can be enabled for network elements. QoS standards may include IEEE 802.1p/Q, the IETF’s RSVP, and DiffServ. Analysis of implementation cost, complexity, management, and benefit must be weighed carefully before adopting such QoS standards. These standards can be recommended when the cost for upgrading some network elements is high and the network resources are scarce and heavily loaded.
3.2. VoIP traffic flow and call distribution
Knowing the current telephone call usage or volume of the enterprise is an important step for a successful VoIP deployment. Before embarking on further analysis or planning phases for a VoIP deployment, collecting statistics about of the present call volume and profiles is essential. Sources of such information are organization’s PBX, telephone records
VIII
and bills. Key characteristics of existing calls can include the number of calls, number of concurrent calls, time, duration, etc. It is important to determine the locations of the call endpoints, i.e. the sources and destinations, as well as their corresponding path or flow. This will aid in identifying the call distribution and the calls made internally or externally. Call distribution must include percentage of calls within and outside of a floor, building, department, or organization. As a good capacity planning measure, it is recommended to base the VoIP call distribution on the busy hour traffic of phone calls for the busiest day of a week or a month. This will ensure support of the calls at all times with high QoS for all VoIP calls. When such current statistics are combined with the projected extra calls, we can predict the worst-case VoIP traffic load to be introduced to the existing network.
Fig. 4 describes the call distribution for the enterprise under study based on the worst busy hour and the projected future growth of VoIP calls. In the figure, the call distribution is described as a probability tree. It is also possible to describe it as a probability matrix. Some important observations can be made about the voice traffic flow for inter-floor and external calls. For all these type of calls, the voice traffic has to be always routed through the router. This is so because Switchs 1 and 2 are layer 2 switches with VLANs configuration. One can observe that the traffic flow for inter-floor calls between Floors 1 and 2 imposes twice the load on Switch 1, as the traffic has to pass through the switch to the router and back to the
IX
switch again. Similarly, Switch 2 experiences twice the load for external calls from/to Floor 3. 3.3. Define performance thresholds and growth capacity
In this step, we define the network performance thresholds or operational points for a number of important key network elements. These thresholds are to be considered when deploying the new service. The benefit is twofold. First, the requirements of the new service to be deployed are satisfied. Second, adding the new service leaves the network healthy and susceptible to future growth. Two important performance criteria are to be taken into account. First is the maximum tolerable end-to-end delay; and second is the utilization bounds or thresholds of network resources. The maximum tolerable end-to-end delay is determined by the most sensitive application to run on the network. In our case, it is 150 ms end-to-end for VoIP. It is imperative to note that if the network has certain delay sensitive applications, the delay for these applications should be monitored, when introducing VoIP traffic, such that they do not exceed their required maximum values. As for the utilization bounds for network resources, such bounds or thresholds are determined by factors such as current utilization, future plans, and foreseen growth of the network. Proper resource and capacity planning is crucial. Savvy network engineers must deploy new services with scalability in mind, and ascertain that the network will yield acceptable performance under heavy and peak loads, with no packet loss. VoIP requires almost no packet loss. In literature, 0.1–5% packet loss was generally asserted. However, in [24] the required VoIP packet loss was conservatively suggested to be less than 10?5. A more practical packet loss, based on experimentation, of below 1% was required in [22]. Hence, it is extremely important not to utilize fully the network resources. As rule-of-thumb guideline for switched fast full-duplex Ethernet, the average utilization limit of links should be 190%, and for switched shared fast Ethernet, the average limit of links should be 85% [25]. The projected growth in users, network services, business, etc. must be all taken into consideration to extrapolate the required growth capacity or the future growth factor. In our study, we will ascertain that 25% of the available network capacity is reserved for future growth and expansion. For simplicity, we will apply this evenly to all network resources of the router, switches, and switched-Ethernet links. However, keep
X
英文文献 科技类 原文及翻译（电子 电气 自动化 通信）1
