Introduction
We are studying a new VoD system that can provide TV programs to viewers over a network. The objective is to let a viewer watch his or her favorite TV program before or after it is broadcast to assist conventional TV services. This system is different from the current closed VoD system that works over cable networks or in hotels in that it uses a global network architecture, such as the Internet's, to provide TV programs to many viewers throughout the country.

VoD system architecture
The VoD system will have a hierarchical structure similar to that of the Internet. It will be necessary to allocate proxies for temporal storage of the requested programs and to use Virtual Private Networks (VPN) to assure a sufficient Quality of Service (QoS) throughout the network. The VoD system architecture shown in Figure 1 is composed of an archive server, proxies and set top units (STU).

Figure 1: Example of VoD system architecture

Simulation

a. Simulation software
We developed a piece of simulation software with the structure shown in Figure 2. It works over OPNET, which is the most popular network simulation program and finds the optimal resource allocation.
The software will be used to find the optimal allocation of proxies, their buffer sizes, assignment of bandwidth, and other resources in order to get the desired QoS expressed in terms of parameters such as delay and packet loss.

The software has the following features:

- Real-time Transport Protocol (RTP) for transmission

- Real-Time Streaming Protocol (RTSP) for control

- User Access Model created from actual statistics

- Program handlers at the server for virtual I/O

- Algorithms (cache replacement, cache routing, Call Admission Control (CAC), etc.) can be added.

This software can evaluate the following items.

i.	VPN traffic profile that includes the VPN trunk traffic parameters and the requirements for packet loss and delay.
ii.	CAC & QoS. CAC is a request admission control mechanism for using resources effectively. QOS is the quality provided by the network. It includes service blocking probability (e.g. 1%) and service response delay.
iii.	Cache principle. The program distribution between the server and proxy sites is very important. The capacity of the cache buffer is limited; hence, the usage of the cache, such as the decision as to which files to delete from the cache, need to be investigated.
iv.	RTP profile. This includes the packet size and the synchronization for the RTP transmission. It also includes the transmission mode.

Figure 2: Structure of the simulation software for the new VoD

b. Results
(1) Effect of proxy
During the simulation, the STU behavior is approximated by the user access model, where the request access rate can be assumed to be a Poisson distribution and the user access interval is approximated by a pre-defined probability distribution obtained from data provided by NHK.
The content preference metric is approximated by using a probability distribution from video shop statistics. The proxy cache performance is then simulated for various user parameters.
Figure 3 shows the access rate's effect on the number of sessions over the link between the proxy and the archive. The number of sessions over the link depends heavily on the access behavior and the number of program handlers at the proxy's site. Although this is only a provisional result given the limited video resources, it is clear that this simulation software can be used to predicate the system parameters if it is given actual video resource data.

Figure 3: Access rate's effect on the number of sessions

(2) CAC
CAC decides whether a connection request can be accepted or rejected. We studied the Measurement Based Call Admission (MBCA) control method with this software. MBCA judges whether it is possible to permit a new connection by using the current packet delay and the traffic parameters to predict the packet delay after adding the new connection. Until now, these packet delays were calculated using an equation based on the node analytic model.
Figure 4 shows lambda plotted against packet length. Lambda is the ratio of the simulated value to the value calculated by the equation.
The results for the simulation and the calculation are almost equal when packet length is long. However, the value calculated by the equation is too large in comparison with the simulation value when packet length is short.
Thus, this simulation software predicts the system performance more precisely in the case of actual packet sizes.

Figure 4: Result of MCBA simulation

Conclusion
The results of the current simulation are preliminary. Still, it can analyze the performance of a VoD system given practical data. The simulation software will be expanded by adding various optional blocks to give a more accurate and realistic simulation. These blocks include algorithms for cache replacement, cache routing, and multicasting.


(Wang Kaiming, SARFT (the State Administration of Radio, Film, and Television) of China)