How to implement QoS filtering in real-time video multicast

1 Introduction
With the development of multicast technology and the popularization of distributed multimedia applications, many real-time video applications based on IP multicast have emerged on the Internet. However, these applications face a major problem: how to implement QoS Filtering Based on TCP-friendly TCPfriendliness in a heterogeneous and dynamic network environment. In a large real-time video multicast system, QoS filtering determines the overall performance of the system.

The Network-Based QoS filtering method is implemented on the gateway or network node, but the QoS filtering method is more feasible because of the Data encapsulation method in the protocol data unit PDU. In the Receiverdriven Layered Multicast (Receiverdriven) [1, 2] QoS filtering method (RLM), the source end divides the video stream into the basic layer and several enhancement layers, through Independent Multicast Group Publishing, the receiving end automatically increases or decreases the number of multicast groups added based on the available bandwidth of the network. This method can improve the QoS of the receiver to a certain extent, but the fixed video hierarchy and layered rate cannot automatically adapt to the dynamic network environment, thus affecting the effect of QoS filtering. To solve this problem, this paper proposes a bidirectional Sender and Receiverdriven Layered Multicast Based on the sending and receiving ends. The QoS filtering method is abbreviated as SRLM ). This method makes full use of the hierarchical and adjustable bit rate encoding technology of video coding standards such as MPEG-4 and H.264, and based on closed loop feedback, it can significantly improve the effect of QoS filtering.

2 method implementation principle
SRLM constructs a Bidirectional Feedback Channel between the source and the acceptor so that the source can use the feedback channel to collect the bandwidth demand information of the acceptor, the optimization algorithm is used to dynamically adjust the video layers and encoding rates at different layers. At the same time, the adaptive results can be actively sent to the receiving end through the feed-forward channel in the form of multi-broadcast messages. After receiving packets, each receiver can use this as a basis for adjustment. Based on its own effective bandwidth, it can decide to increase or decrease or maintain the number of reserved video layers. In RLM, in order to maximize the receiving quality, the receiver must introduce a coordination or knowledge sharing mechanism [1] to eliminate network congestion caused by blind attempts to increase video layering.

2.1SRLM source implementation algorithm

For ease of description, make the following settings:

1) The source end adopts the cumulative hierarchy method, and the maximum hierarchy constraint is limit L. The first layer is the basic layer, followed by the enhancement layers.

2) bi corresponds to the bit rate at each layer, I = ,..., L. The bitrate constraint for each layer is bi ε bimin ,..., Bimax, which is a finite number of discrete values. Set ci = Σ ij = 1bj, j = 1, 2 ,..., I (I ≤ L), ci is the layer-by-layer bitrate accumulation from Layer 1 to layer I and ci-ci-1 = bi. Convert β l = c1, c2 ,..., Cl is defined as a layer-by-layer cumulative bit rate vector. l ≤ L. c1 = b1 is the lower limit of the encoding rate of the basic layer.

3) set the number of users joining multicast sessions to N, and the expected receiving rate set to r1, r2, r3 ,..., RN, now we introduce the receiver's Equity Index definition: Jri, β l = Gini ri, β lri, I ≤ N, where Gini ri, β l = maxc: c ≤ ri, c, c1, c2 ,..., Cl, which is the current valid receiving bit rate of the receiver I. Obviously, Jri, β l is less than or equal to 1. Make Jr, β l = 1N Σ Ni = 1Jri, and β l is the overall acceptance fairness index of N receivers.

According to the preceding settings, source QoS filtering can be defined as: how to determine the maximum number of layers l' and layer-by-layer cumulative bit rate vector β l' under the constraints of conditions 1) and 2 ′, use Jr and βl' to obtain the maximum value.

The algorithm is described as follows:

1) collect RR feedback packets and disassemble the packets to obtain the sets of r1, r2, r3 ,..., RN, and calculates the cumulative bit rate vector β l = c1, c2 ,..., The overall acceptance Equity Index under cl.

2) According to the constraints of conditions 1) and 2), all the ri with a number greater than Σ Li = 1bimax are combined into a value equal to Σ Li = 1bimax, if the bandwidth is lower than b1min, the deletion is not included in the calculation of the average equity index. In this way, the number of items that the receiver effectively expects to have repeated items in the receiving rate set is still N '= N-m-n + 1, where m is the number of ri items whose value exceeds Σ Li = 1bimax, and n is the number of ri items whose value is lower than b1min. Then, we can obtain the pre-processed receiver's desired receiving rate set r'1, r'2 ,..., R'n ′.

3) change the formula Jr, β l = 1N Σ Ni = 1Jri, and β l to Jr ', β L' = 1n' Σ n'' I = 1Jr' I, βl', which is used as the data to calculate the test and search for the L' And βl' that bring the overall Equity Index to the maximum value ′. Defines the adjustment margin ε. When Jr ', β l'-Jr, β l> ε, the encoder makes corresponding optimization adjustments according to the parameters l' and β l, otherwise, the current status is maintained to ensure the relative stability of the system.

4) return the first step to start again. When the source side executes the preceding steps, it sends an SR packet to each receiver every second in the TSR multicast mode to report the current number of video layers at the source end and layer-by-layer cumulative bit rate vector β l. TSR is set to: TSR = TAdj/k, k is an integer greater than 1, TAdj is the cycle of the source side to execute the above algorithm, also known as the source side adjustment cycle. The selection of TAdj cannot be too small. Otherwise, the computing burden on the large source end will be added and the adjustment and oscillation will be caused, which is not conducive to the transmission and receipt of real-time video data, it also makes it difficult to correctly collect and calculate the expected receiving rate reports of each receiver. The analysis shows that the value of TAdj is 10 ~ 15 s is reasonable.

2.2SRLM receiver Algorithm Implementation

The acceptor algorithm is described as follows:

1) estimate the values of p, tRTT, and tRTO parameters;

2) If the SR message is received, go to the next step; otherwise, go back to the previous step;

3) disassemble the SR packet to obtain the current β l of the source end. 1) Calculate the expected receiving rate r of the local receiving end;

4) Adjust the number of video layers subscribed at the local end based on β l and expected receiving rate r.

5) return the first step.

The receiving end sends an RR packet to the source end every TRR second while performing the preceding steps to report the expected receiving rate of the receiving end. The report is also used as a request for tRTT estimation.

TRTT uses a closed-loop estimation method. When receiving the RR packet from the receiving end, the source end returns an SR packet as a response. To reduce the system overhead, the source end does not separately respond to each RR packet, but processes it in batches. That is, assume that the source side sends an SR packet at the local t time, And the treturn1, treturn2 ,..., Treturnk receives RR packets from k receivers one after another, all of which have their own synchronization source identifier SSRCi. The source end sends the next SR packet in the t + TSR multicast mode at the moment. In addition to the β l parameter, the packet also contains a list, the list contains the SSRCi of the above k receivers and its corresponding tdelayi. Tdelayi is the delay time for The Source end to respond to a specific RR packet, that is, the interval between the source end receiving a packet and sending the next SR packet ).

We can see that tdelayi = t + TSR-treturni. After receiving the SR message, if the receiving end searches for the SSRCi → tdelayi corresponding to the local end, then, the closed loop estimation value of tRTT can be obtained through Tau 0 = t'-tdelayi. The local time when the receiving end sends the RR message and receives the SR message, respectively. When the receiving end sends an RR packet after TSR + tRTO (seconds) but does not receive the corresponding SR packet response, it is deemed that the SR packet has been lost, and the receiving end clears the request record and sends a new RR packet.

The tRTO parameter can be calculated by using tRTO = max1, 4 Tau 0 [4]. This empirical formula can provide good TCP Friendliness in practical application. The estimation of Packet Loss Rate p can be referred to in [4]. However, when the number of video layers received by the receiver is greater than 1, it is necessary to treat each layer of video streams separately, therefore, the total packet loss rate p is calculated based on the packet loss rate of each layer.

3. Simulation Experiment and Performance Evaluation
In this section, the performance of SRLM is simulated and analyzed through the network simulator NS-2 [5.

The network topology established in the simulation experiment is shown in 3, the following default settings are used: 1) the FIFO/droptail scheduling strategy is adopted, the maximum queuing delay is 150 MS; 2) the transmission delay between routers is 20 MS, the transmission delay between the router and the terminal is 10 MS; 3) the TCP stream continues throughout the simulation experiment, and the maximum transmission window is 4 000 packets; 4) multicast Video Stream and TCP packet size are unified to 500 bytes; 5) the initial parameters of the Video Receiver is set to: tRTT = 100 MS, p = 0.

In the experiment of verifying TCP stream friendliness of SRLM, the initial rate of each layer of the source video is set to 256,512,102 4 kbps, and the video receiving end in the LAN is set to 5.

SRLM enables TCP stream and video multicast stream to better share the link bottleneck and maintain similar transmission bandwidth, thus demonstrating good fairness.

In order to further verify the effectiveness of SRLM, it is also compared with RLM in the overall acceptance fairness index in the simulation experiment. The basic settings of the experiment are as follows:

1) RLM two cumulative layer Bit Rate Distribution Method: a) uniform Distribution Type uniform allocation, figure 5 referred to as RLM-UA), that is, the ci = ci-1 + σ, in the experiment, σ = (2 048-128)/L; B) exponential allocation, which is referred to by RLM-EA in Figure 5), that is, the ci = λ ci-1, in the experiment, λ = L2048/128. A) and B) Indicate the number of layers. The minimum rate of the basic layer is set to 128 kbps.

2) the bit rate of the accumulation layer of SRLM is the same as that of B). Each layer has 128/L integers.) the rate adjustment points are fixed at a constant speed.

3) The simulation environment is 1 000 multicast receivers, which are distributed in a cluster according to the predefined number of video layers, we apply the Gaussian Mixture Distribution Model to the expected receiving rate ri of each receiver [6], and assume there are five clusters, each of which has 200 receivers ), the average receiving rate set of clusters is 160,360,800, 800.

The comparison results reflect the QoS filtering performance of SRLM. We can also see that when the video has more than five layers, the overall system receives the fairness index, and the excessive number of layers will increase the computing workload and complexity at both ends, instead, it will reduce the QoS filtering effect of real-time video multicast to a certain extent. In actual application, 3 ~ The layer 5 has been able to adapt to the number of multicast receiving sessions.

4 Conclusion
In the current dynamic and heterogeneous IP networks, the video QoS Filtering Method Based on the terminal system is more feasible. Through comprehensive application of various congestion and rate control policies and error code control mechanisms in the terminal system, we can provide certain QoS guarantees for real-time multicast videos, thus improving the overall quality of video stream reception. However, with the development and progress of network technology, the network-based QoS filtering method will gradually mature and become the main application method.

Related Articles]

• IP multicast technology and Programming

• Implement CDN publishing network bandwidth management and QoS implementation

• One Huawei router configuration that uses multicast and Qos