A method for virtualizing audio integrated network resources

Abstract

The application discloses an audio integrated network resource virtualization method, and belongs to the technical field of audio network resource management, and comprises the following steps: constructing a distributed virtual switching layer of a covering subnet, embedding a virtual switch instance into a kernel state network stack of each physical host in a cluster, and dynamically mapping cloud hosts in the subnet to the physical host group through a consistent hashing algorithm; when the cloud host sends an audio data packet, the source physical host virtual switch instance performs the following steps: extracting a DSCP field and marking it as EF, encapsulating through a GRE-over-UDP tunnel, and replacing a target MAC address with a virtual MAC address of a next-hop virtual switch instance; after the virtual switch instance of the target physical host is unpacked, the audio packet is delivered to the target cloud host based on a virtual port forwarding table. Through the distributed virtual switching layer and the consistent hashing mapping mechanism, the method realizes efficient virtualization transmission of audio data, supports QoS guarantee and low-delay transmission, and is suitable for audio network resource management in a large-scale cloud environment.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of audio network resource management, and particularly relates to a virtualization method for audio integrated network resources. BACKGROUND

[0002] With the rapid development of cloud computing and virtualization technology, the real-time transmission demand for audio streams (such as voice calls, video conferences, etc.) in the cloud environment is increasing.

[0003] Traditional virtualization networks usually use centralized virtual switches or virtual network devices to manage network traffic in the cloud environment. This architecture has the following problems when processing audio streams: the traditional virtualization network resource allocation strategy is usually static, and cannot dynamically adjust the bandwidth, buffer size and other resource parameters of the virtual port according to the real-time audio stream demand, resulting in low resource utilization or performance bottlenecks; the traditional virtualization network lacks a special recognition and optimization mechanism for audio streams, and audio streams are mixed with other types of data streams (such as video streams, ordinary data packets) for transmission, which can easily cause delay, jitter and packet loss problems of audio streams, affecting user experience; the traditional virtualization network architecture usually relies on complex physical network devices (such as physical switches, routers, etc.) and virtual network devices (such as virtual switches, virtual firewalls, etc.), and this multi-layer architecture can increase network latency and management complexity in large-scale cloud environments.

[0004] Therefore, it is necessary to propose a virtualization method for audio integrated network resources to solve the above problems. SUMMARY

[0005] Therefore, the purpose of the present application is to provide a virtualization method for audio integrated network resources, which realizes efficient transmission of audio streams and dynamic optimization of resources through the cooperative work of audio stream recognition, virtual port manager and policy execution engine.

[0006] To achieve the above purpose, the present application provides the following technical scheme:

[0007] The present application provides a virtualization method for audio integrated network resources, comprising the following steps: S1: constructing a distributed virtual switch layer covering a subnetwork, embedding a virtual switch instance into the kernel mode network stack of each physical host in the cluster, the virtual switch instance comprising an audio stream recognition module, a virtual port manager and a policy execution engine;

[0008] S2: dynamically mapping the cloud hosts in the subnetwork to the physical host group through a consistent hash algorithm, and the mapping relationship satisfies:

[0009]

[0010] In the formula, The SHA-256 hash function is used to map input to a fixed-length hash value. A unique identifier for the virtual machine. A unique identifier for the physical host. For bitwise XOR operation, For the preset threshold, ;

[0011] S3: When the cloud host sends audio data packets, the source physical host virtual switch instance performs the following: extracts the DSCP field and marks it as EF, encapsulates it through a GRE-over-UDP tunnel, and replaces the target MAC address with the virtual MAC address of the next-hop virtual switch instance;

[0012] S4: After the virtual switch instance of the target physical host is decapsulated, the audio packet is delivered to the target cloud host based on the virtual port forwarding table.

[0013] Furthermore, the audio stream recognition module employs a dual-threshold triggering mechanism, in which...

[0014] The first threshold is: the size of 5 consecutive packet payloads. Bytes and packet spacing ;

[0015] The second threshold is: RTP sequence number increment. deviation rate ;

[0016]

[0017] In the formula, This is the actual serial number increment. This is the theoretical sequence number increment. ;

[0018] When both the first and second thresholds are met, the dedicated audio forwarding channel is activated, and the deviation rate... If the signal is detected, it is determined to be a discontinuous audio stream or a network attack stream, and the security policy is triggered: the data packet is dropped and an alarm event is generated.

[0019] Furthermore, the execution steps of the dedicated audio forwarding channel include:

[0020] A1: Bypass the traditional Linux TC flow table and directly call the eBPF program for forwarding;

[0021] A2: Allocate an independent memory pool;

[0022] A3: Enable forward error correction coding: ,

[0023] in =4, For forward error correction block, For continuous XOR operation from the first to the kth data packet, For the i Original audio data packet.

[0024] Further, in step A3, the forward error correction coding adopts a non-uniform protection mechanism, including:

[0025] When the silence detection flag is False, it is determined as a key audio frame, and the key audio frame is allocated according to the redundancy ; In the formula, The current path packet loss rate is

[0026] When the silence detection flag is True, it is determined as a non-key audio frame, and the non-key audio frame is allocated according to the redundancy .

[0027] Further, the optimization strategy executed by the policy execution engine includes: monitoring the audio stream peak bandwidth in the cluster in real time , when the bandwidth ,

[0028] Start zero-copy buffer expansion:

[0029]

[0030] Trigger vCPU affinity rebinding, bind the interrupt thread processing the audio stream to the physical core P-core;

[0031] Dynamically adjust the GRE tunnel MTU value:

[0032]

[0033] In the formula, The reference bandwidth is The buffer size after expansion, The buffer size before expansion, The total bandwidth capacity, The adjusted Value.

[0034] Further, the steps of GRE-over-UDP tunnel encapsulation include:

[0035] B1: Establish a compression dictionary , including source IP, destination IP, GRE Key, UDP port;

[0036] B2: Difference encoding and Huffman encoding of the packet header by the following compression algorithm,

[0037]

[0038] wherein, is the compressed header information, is the header information of the current data packet, is the dictionary database, is the dictionary reference index, is the Huffman encoding operation, is the difference extraction operator, is the data connector;

[0039] B3: When the audio stream duration T is detected for more than 2s, update the compression dictionary:

[0040]

[0041] wherein, is the dictionary update smoothing factor, taken , is the updated dictionary, is the dictionary before update.

[0042] Further, in step S4, the virtual port forwarding table is updated by the voiceprint-based Qos scheduling algorithm, the updating step is:

[0043] C1: Extract the MFCC feature vector of the audio stream ;

[0044] C2: Calculate the cosine similarity of the current audio stream and the high-priority audio stream :

[0045]

[0046] wherein, , is the i-th component of the MFCC feature vector of the current audio stream, is the i-th component of the MFCC feature vector of the high-priority audio stream; C3: When ,

[0047] assign Qos weight according to the similarity interval :

[0048] and schedule the priority

[0049] .

[0050] Further, in step S3, the target MAC replacement adopts network situation-based path selection, and selects the path quality and the hop count​​ The path is: wherein,

[0051]

[0052] In the formula, , , The weight factor is 0.5, 0.3, 0.2, respectively, The normalized path delay is, The path packet loss rate is, The path jitter is.

[0053] The beneficial effects of the present application are:

[0054] The present application deeply couples the distributed switching architecture with the audio service characteristics, breaks through the delay bottleneck of the traditional centralized architecture through the kernel state embedded vSwitch + audio stream feature recognition + dynamic hash mapping, reduces the end-to-end delay to below 50ms, the voice packet loss rate is <0.3%, and supports >1000 node elastic expansion.

[0055] Other advantages, objects, and features of the present application will be set forth in the following specification and will be apparent to those skilled in the art from the teachings of the present application. The objectives and other advantages of the present application will be realized and attained by the structure particularly pointed out in the written description and claims thereof as well as the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to make the purposes, technical solutions and beneficial effects of the present application clearer, the present application provides the following drawings for illustration:

[0057] Figure 1 The flowchart of the embodiment of the present application. DETAILED DESCRIPTION

[0058] As Figure 1 shown, the present application provides a virtualization method of audio integrated network resources, comprising the following steps:

[0059] S1: Construct a distributed virtual switching layer covering a subnet, embed a virtual switch instance (vSwitch Instance) in the kernel state network stack of each physical host in the cluster, the virtual switch instance includes a sound stream frequency recognition module, a virtual port manager and a policy execution engine;

[0060] Among them, the sound stream frequency recognition module identifies the audio stream data packet based on the Payload Type field of the real-time transport protocol (RTP) header and the feature that the packet length is ≤250 bytes;

[0061] The virtual port manager supports 802.1Q VLAN tagging, and is responsible for dynamically adjusting resource allocation and performance parameters of the virtual port;

[0062] The policy execution engine stores the ACL rules and QoS queues, and is responsible for executing the predefined optimization strategy according to the real-time system state;

[0063] S2: The cloud hosts in the subnet are dynamically mapped to the physical host group through a consistent hashing algorithm, and the mapping relationship satisfies:

[0064]

[0065] In the formula, is a SHA-256 hash function for mapping input to a fixed-length hash value, is a unique identifier of the virtual machine, is a unique identifier of the physical host, is a bitwise XOR operation, is a preset threshold, ;

[0066] S3: When the cloud host sends an audio data packet, the source physical host virtual switch instance performs: extracts the DSCP field and marks it as EF, encapsulates through GRE-over-UDP tunneling, and replaces the target MAC address with the virtual MAC address of the next-hop virtual switch instance;

[0067] Wherein, the target MAC address replacement adopts network situation-based path selection, and selects a path with path quality and hop count ; wherein,

[0068]

[0069] In the formula, , , is a weight factor, which is 0.5, 0.3, and 0.2 respectively, is the normalized path delay, is the path packet loss rate, is the path jitter;

[0070] S4: After the target physical host's virtual switch instance decapsulates, it delivers the audio packet to the target cloud host based on the virtual port forwarding table.

[0071] In the present scheme, the distributed switching architecture is deeply coupled with the audio service characteristics, and through kernel-embedded vSwitch + audio stream feature recognition + dynamic hashing mapping, the delay bottleneck of traditional centralized architecture is broken through; the end-to-end delay is reduced to below 50ms, the voice packet loss rate is <0.3%, and >1000-node elastic expansion is supported.

[0072] In one embodiment of the present invention, the audio stream identification module employs a dual-threshold triggering mechanism:

[0073] First threshold: 5 consecutive packet payload sizes Bytes and packet spacing ;

[0074] Second threshold: RTP sequence number increment deviation rate satisfy:

[0075]

[0076] In the formula, This is the actual serial number increment. This is the theoretical sequence number increment. ;

[0077] When both thresholds are met simultaneously, the dedicated audio forwarding channel is activated, and this channel performs the following:

[0078] A1: Bypass the traditional Linux TC flow table and directly call the eBPF program for forwarding;

[0079] A2: Allocate an independent memory pool;

[0080] A3: Enable forward error correction coding: ,

[0081] in =4, For forward error correction blocks, To perform a continuous XOR operation from the 1st to the kth data packet, For the first i One raw audio data packet;

[0082] If the deviation rate If the signal is detected, it is determined to be a discontinuous audio stream or a network attack stream, and the security policy is triggered: the data packet is dropped and an alarm event is generated.

[0083] This solution constructs a dual verification model based on the spatiotemporal characteristics of audio packets and combines eBPF kernel bypass technology to solve the jitter sensitivity problem.

[0084] In one embodiment of the present invention, the forward error correction coding employs a non-uniform protection mechanism to improve the packet loss resistance of key audio segments. The non-uniform protection mechanism includes:

[0085] When the silence detection flag is False, it is determined to be a critical audio frame, and the critical audio frames are sorted according to redundancy. Distribute; where, This represents the packet loss rate for the current path.

[0086] When the mute detection flag is True, it is determined that the non-key audio frame is assigned according to the redundancy of the non-key audio frame.

[0087] In an embodiment of the present application, the optimization strategy executed by the policy execution engine includes: monitoring the audio stream peak bandwidth in the cluster in real time , when the bandwidth ,

[0088] Starting zero-copy buffer expansion:

[0089]

[0090] Triggering vCPU affinity rebinding, binding the interrupt thread processing the audio stream to the physical core P-core;

[0091] Dynamically adjusting the GRE tunnel MTU value:

[0092]

[0093] In the formula, is the reference bandwidth, is the size of the expanded buffer, is the size of the buffer before expansion, is the total bandwidth capacity, is the adjusted value.

[0094] The scheme directly associates network resource allocation with audio service load, realizes sub-second level elastic response, reduces voice interruption time under burst traffic, and improves resource utilization.

[0095] In an embodiment of the present application, the steps of GRE-over-UDP tunnel encapsulation include:

[0096] B1: Establishing a compression dictionary , including source IP, destination IP, GRE Key, and UDP port;

[0097] B2: Difference encoding and Huffman encoding are performed on the data packet header through the following compression algorithm,

[0098]

[0099] In the formula, is the compressed header information, is the header information of the current data packet, is the dictionary database, is the dictionary reference index, is the Huffman encoding operation, For difference extraction operators, For data concatenation;

[0100] B3: When the duration T of the detected audio stream is... Update the compression dictionary at 2 seconds:

[0101]

[0102] In the formula, Update the smoothing factor for the dictionary, take , For the updated dictionary, This is the dictionary before the update.

[0103] This solution is designed with an incremental dictionary update mechanism to address the long-term connection characteristics of audio streams. It breaks through the limitations of the fixed dictionary in the traditional ROHC protocol, compressing the tunnel header overhead from 40 bytes to an average of 8 bytes, and improving bandwidth utilization by 25%.

[0104] In one embodiment of the present invention, in step S4, the virtual port forwarding table is updated using a voiceprint-based QoS scheduling algorithm. The update step is as follows:

[0105] C1: Extract the MFCC feature vector of the audio stream. ;

[0106] C2: Calculate the cosine similarity between the current audio stream and the higher-priority audio stream. :

[0107]

[0108] In the formula, , The MFCC eigenvector of the current audio stream is the first... One portion, The first MFCC eigenvector of the high-priority audio stream One component;

[0109] C3: When At that time, Qos weights are assigned based on similarity intervals. :

[0110]

[0111] and scheduling priority .

[0112] This solution prioritizes key speech streams, thus improving the MOS score.

[0113] Finally, it should be noted that the above preferred embodiments are merely intended to illustrate the technical solutions of the present application but not to limit the present application. Although the present application has been described in detail through the above preferred embodiments, those skilled in the art should understand that various modifications can be made in form and details without departing from the scope of the present application defined by the claims.

Claims

1. A method for virtualization of audio integrated network resources, characterized by, The method comprises the following steps: S1: Construct a distributed virtual switch layer covering the subnets, embed a virtual switch instance into the kernel-mode network stack of each physical host in the cluster, and the virtual switch instance comprises an audio stream identification module, a virtual port manager, and a policy execution engine; S2: Dynamically map the cloud hosts in the subnet to the physical host group through a consistent hashing algorithm, and the mapping relationship satisfies: In the formula, SHA-256 is a hash function for mapping input to a fixed-length hash value, is a unique identifier of a virtual machine, is a unique identifier of a physical host, is a bitwise XOR operation, is a preset threshold value, ; S3: When the cloud host sends an audio data packet, the source physical host virtual switch instance performs: extracts the DSCP field and marks it as EF, encapsulates through a GRE-over-UDP tunnel, and replaces the target MAC address with the virtual MAC address of the next-hop virtual switch instance; S4: After the virtual switch instance of the target physical host decapsulates, the audio packet is delivered to the target cloud host based on the virtual port forwarding table; The audio stream identification module adopts a double-threshold triggering mechanism, wherein, The first threshold is: 5 consecutive packet payload sizes Bytes and packet interval ; The second threshold is: RTP sequence number increment of the bias rate <0.1 ; In the formula, is the actual sequence number increment, is the theoretical sequence number increment, ; When both the first threshold and the second threshold are satisfied, the audio dedicated forwarding channel is activated, when the deviation rate then it is determined as a non-continuous audio stream or a network attack stream, and a security policy is triggered: discarding the data packet and generating an alarm event.

2. The method of virtualization of audio integrated network resources according to claim 1, characterized in that: The steps performed by the audio dedicated forwarding channel include: A1: Bypass the traditional Linux TC flow table and directly call the eBPF program for forwarding; A2: Allocate an independent memory pool; A3: Enable forward error correction coding: , wherein = 4, is a forward error correction block, is a continuous XOR operation from the 1st to the kth data packet, is the i original audio data packet.

3. The method of virtualization of audio integrated network resources according to claim 2, characterized in that: In step A3, the forward error correction coding adopts a non-uniform protection mechanism, comprising: When the silence detection flag is False, the critical audio frame is determined, and the critical audio frame is allocated according to the redundancy ; where, is the current path packet loss rate; When the silence detection flag is True, it is determined that the audio frame is a non-key frame, and the non-key audio frame is distributed according to the redundancy assigned.

4. The method of virtualization of audio integrated network resources according to claim 1, characterized in that: The optimization strategy executed by the policy execution engine includes: monitoring the audio stream peak bandwidth in the cluster in real time When the bandwidth is Start the zero-copy buffer expansion: Trigger vCPU affinity rebinding to bind the interrupt thread processing the audio stream to the physical core P-core; Dynamically adjust the GRE tunnel MTU value: In the formula, is the reference bandwidth, is the expanded buffer size, is the buffer size before expansion, is the total bandwidth capacity, is the adjusted value.

5. The method of virtualization of audio integrated network resources according to claim 1, characterized in that: The steps of the GRE-over-UDP tunnel encapsulation include: B1 : Establishing a compression dictionary including source IP, destination IP, GRE Key, UDP port; B2: Perform differential encoding and Huffman encoding on the packet header through the following compression algorithm, wherein, is the compressed header information, is the header information of the current data packet, is the dictionary database, is the dictionary reference index, is the Huffman encoding operation, is the difference extraction operator, is the data connector; B3: When a sound stream is detected for a duration T 2s, update the compression dictionary: In the formula, is a dictionary update smoothing factor, taken as , is the updated dictionary, is the dictionary before updating.

6. The method of virtualization of audio integrated network resources according to claim 1, characterized in that: In step S4, the virtual port forwarding table is updated through a voiceprint-based Qos scheduling algorithm, and the updating steps are: C1 : Extracting MFCC feature vectors of the audio stream ; C2: calculating a cosine similarity of the current audio stream and the high priority audio stream : In the formula, , The MFCC eigenvector of the current audio stream is the first... One portion, The first MFCC eigenvector of the high-priority audio stream One component; C3: When the Qos weight is assigned according to the similarity interval : And dispatching priority .

7. The method of virtualization of audio integrated network resources according to claim 1, characterized in that: In step S3, the target MAC replaces the network situation-based path selection and selects a path with path quality and hop count ; wherein, wherein, , , are weight factors, respectively taken as 0.5, 0.3, 0.2, is the normalized path delay, is the path packet loss rate, is the path jitter.

Images