Network scheduling method and system for hybrid service transmission based on RDMA (Remote Direct Memory Access)

Abstract

The invention is suitable for the technical field of network scheduling, and provides a network scheduling method and system for hybrid service transmission based on RDMA, and the method comprises the steps: receiving a to-be-transmitted service demand instruction, and determining a corresponding service type according to a service quality parameter in the service demand instruction; matching a corresponding virtual transmission channel for the service demand instruction according to the service type, wherein the virtual transmission channel is used for isolating hardware-level resources of different types of traffic; selecting an optimal transmission path for the flow of each virtual transmission channel according to a preset path evaluation model; analyzing the congestion risk of the optimal transmission path by adopting a prediction model; and dynamically adjusting a flow transmission path according to a congestion risk analysis result, and ensuring the transmission quality of the highest priority service through a hardware-level scheduling strategy. According to the invention, the optimal balance between the service quality and the resource efficiency can be realized in a complex mixed service environment.

Description

Technical Field

[0001] The present invention relates to the technical field of network scheduling, and in particular to a network scheduling method and system based on RDMA-based hybrid service transmission. Background Art

[0002] In recent years, with the rapid development of cloud computing, big data, and artificial intelligence technologies, the types of services carried by data center networks have become highly diverse and complex, forming a hybrid business scenario that includes real-time interactive services, high-throughput services, and low-latency sensitive services. Existing RDMA-based network scheduling methods lack the ability to perceive multi-dimensional business characteristics when facing the transmission needs of hybrid services. Currently, the scheduler only performs coarse-grained classification based on basic information in the packet header and cannot dynamically identify the real-time transmission characteristics of the service. In addition, current scheduling methods mostly use a fixed-weight resource allocation model that does not take into account the dynamic changes in network status. This can easily lead to problems such as service quality fluctuations, low resource utilization, and increased interference between services when supporting hybrid services in data center networks.

[0003] In view of this, a network scheduling method and system for hybrid service transmission based on RDMA is provided. Summary of the Invention

[0004] The embodiments of the present application provide a network scheduling method and system for hybrid service transmission based on RDMA, which is used to solve the problems of service quality fluctuations, low resource utilization and increased interference between services in data center networks when supporting hybrid services.

[0005] A first aspect of an embodiment of the present application provides a network scheduling method for hybrid service transmission based on RDMA, including:

[0006] receiving a service requirement instruction to be transmitted, and determining a corresponding service type according to a quality of service parameter in the service requirement instruction;

[0007] Matching a corresponding virtual transmission channel to the service demand instruction according to the business type, wherein the virtual transmission channel is used to isolate hardware-level resources of different types of traffic;

[0008] Select the optimal transmission path for the traffic of each virtual transmission channel according to the preset path evaluation model;

[0009] Analyzing the congestion risk of the optimal transmission path using a prediction model;

[0010] Dynamically adjust traffic transmission paths based on congestion risk analysis results, and ensure the transmission quality of the highest priority services through hardware-level scheduling strategies.

[0011] Furthermore, the receiving of the service requirement instruction to be transmitted and determining the corresponding service type according to the quality of service parameter in the service requirement instruction includes:

[0012] If the delay threshold is less than or equal to the preset threshold, the service demand instruction is delay-sensitive;

[0013] If the bandwidth quota is greater than or equal to a first predetermined percentage of the total bandwidth, the service demand instruction is bandwidth-intensive;

[0014] If the packet loss tolerance is less than or equal to the second preset percentage, the service demand instruction is reliability-sensitive;

[0015] If the burst coefficient is greater than or equal to the preset burst threshold, the service demand instruction is a burst service;

[0016] The service demand instruction belongs to one or more service types selected from delay-sensitive, bandwidth-intensive, reliability-sensitive and bursty services.

[0017] Furthermore, the matching of the service demand instruction to a corresponding virtual transmission channel according to the business type, wherein the virtual transmission channel is used to isolate hardware-level resources for different types of traffic, includes:

[0018] If the service demand instruction belongs to a service type, matching a corresponding virtual transmission channel according to the service type through the RNIC network card;

[0019] If the service demand instruction belongs to multiple business types at the same time, the RNIC network card will match the virtual transmission channel corresponding to each business type according to the preset priority rules. The preset priority matching rules are that delay-sensitive type takes precedence over reliability-sensitive type, reliability-sensitive type takes precedence over bandwidth-intensive type, and bandwidth-intensive type takes precedence over bursty business.

[0020] Furthermore, the matching of the service demand instruction to a corresponding virtual transmission channel according to the business type, wherein the virtual transmission channel is used to isolate hardware-level resources for different types of traffic, further includes:

[0021] The delay-sensitive service is matched with a first virtual transmission channel, and the first virtual transmission channel adopts a strict priority queue;

[0022] The reliability-sensitive service is matched to a second virtual transmission channel, and the second virtual transmission channel is a weighted fair queue;

[0023] The bandwidth-intensive service is matched with the third virtual transmission channel, and the third virtual transmission channel adopts the PFC flow control mechanism;

[0024] The bursty service matches the fourth virtual transmission channel, and the fourth virtual transmission channel supports dynamic buffer expansion.

[0025] Furthermore, selecting the optimal transmission path for the traffic of each virtual transmission channel according to the preset path evaluation model includes:

[0026] The SDN controller obtains all available paths in the current network topology and selects the candidate paths that meet the preset conditions. The preset conditions include that the delay of the current path is less than or equal to a preset multiple of the service delay threshold, and the remaining bandwidth of the current path is greater than or equal to half of the service bandwidth quota;

[0027] Collect evaluation parameters of each candidate path in real time, including link delay, remaining bandwidth, packet loss rate and historical path stability;

[0028] The scores of the candidate paths are calculated according to the evaluation parameters, and the path with the highest score is selected as the optimal transmission path.

[0029] Furthermore, calculating the scores of the candidate paths according to the evaluation parameters and selecting the path with the highest score as the optimal transmission path includes:

[0030]

[0031] Where: L is the delay, RB is the remaining bandwidth, PLR is the packet loss rate, HPS is the historical path stability, α, β, γ and δ are the corresponding weight coefficients respectively.

[0032] Furthermore, the adopting of a prediction model to analyze the congestion risk of the optimal transmission path includes:

[0033] Obtaining the data source of the optimal transmission path from the SDN controller and the RNIC;

[0034] The data source generates time series feature data according to a preset time window, and the time series feature data is input into the trained LSTM prediction model;

[0035] Output the congestion probability and congestion risk level of the optimal transmission path within a preset time.

[0036] Furthermore, the method of dynamically adjusting the traffic transmission path based on the congestion risk analysis results and ensuring the transmission quality of the highest priority services through hardware-level scheduling strategies includes:

[0037] When the congestion probability is greater than or equal to a third preset percentage, switching the transmission path from the optimal transmission path to an alternative path, where the alternative path is an available path with a second-ranked path score and a congestion probability less than a third preset percentage;

[0038] When the congestion probability is greater than or equal to a fourth preset percentage and less than a third preset percentage, half of the traffic in the optimal transmission path is migrated to the alternative path.

[0039] Furthermore, the method of dynamically adjusting the traffic transmission path based on the congestion risk analysis results and ensuring the transmission quality of the highest priority services through hardware-level scheduling strategies also includes:

[0040] When the queue depth of the first virtual transmission channel matching the delay-sensitive service is greater than or equal to a fifth preset percentage of the total capacity, suspending traffic transmission of other virtual transmission channels except the first virtual transmission channel through the RNIC hardware scheduler;

[0041] The suspended traffic is cached in an independent backup queue, and traffic transmission on other virtual transmission channels is resumed when the queue depth of the first virtual transmission channel is less than 50%.

[0042] A second aspect of an embodiment of the present application provides a network scheduling system for hybrid service transmission based on RDMA, including:

[0043] A service requirement instruction receiving unit, configured to receive a service requirement instruction to be transmitted and determine a corresponding service type according to a quality of service parameter in the service requirement instruction;

[0044] A virtual transmission channel matching unit, configured to match a corresponding virtual transmission channel to the service demand instruction according to the service type, wherein the virtual transmission channel is used to isolate hardware-level resources for different types of traffic;

[0045] An optimal transmission path selection unit, configured to select an optimal transmission path for the traffic of each virtual transmission channel according to a preset path evaluation model;

[0046] a congestion risk prediction unit, configured to analyze the congestion risk of the optimal transmission path using a prediction model;

[0047] The traffic transmission path adjustment unit is used to dynamically adjust the traffic transmission path according to the congestion risk analysis results, and to ensure the transmission quality of the highest priority services through hardware-level scheduling strategies.

[0048] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0049] The present invention first determines the corresponding service type according to the service quality parameters in the service demand instruction to be transmitted, and matches the corresponding virtual transmission channel for the service demand instruction according to the service type, wherein the virtual transmission channel is used to isolate the hardware-level resources of different types of traffic, effectively solving the resource preemption and interference problems caused by the competition of mixed service traffic; then selects the optimal transmission path for the traffic of each virtual transmission channel according to the preset path evaluation model. The path evaluation model can dynamically select the optimal path according to the needs of different service types, combining real-time data collection, dynamic weight adjustment and hardware-level preemption mechanism; and adopts a prediction model to analyze the congestion risk of the optimal transmission path, dynamically adjusts the traffic transmission path according to the congestion risk analysis results, and ensures the transmission quality of the highest priority service through the hardware-level scheduling strategy. By identifying possible network congestion in advance, the transmission path is adjusted in time to ensure service quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 The figure is a flow chart of an embodiment of a network scheduling method for hybrid service transmission based on RDMA in the present invention. DETAILED DESCRIPTION

[0051] The terms "first," "second," "third," "fourth," and the like (if any) in the specification and claims of the present invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequential sequence. It should be understood that the numbers used in this manner are interchangeable where appropriate so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "corresponding to," and any variations thereof, are intended to cover non-exclusive inclusions, e.g., a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatus.

[0052] Example 1

[0053] The implementation method in this embodiment can be implemented in the system, can be implemented in the server, and can also be implemented in the terminal, without specific limitation. The following will introduce the network scheduling method for hybrid service transmission based on RDMA in this application from the perspective of system implementation. Figure 1 , the method provided in the embodiment of the present application includes the following steps:

[0054] S11 receives the service demand instruction to be transmitted, and determines the corresponding service type according to the service quality parameters in the service demand instruction;

[0055] In this embodiment, after the system receives a service demand instruction triggered by the application layer, it parses the service quality parameters in the service demand instruction. The service quality parameters here are QoS indicator parameters, including delay threshold, bandwidth quota, packet loss tolerance and burst coefficient. The delay threshold is the maximum end-to-end delay allowed by the service, the bandwidth quota is the guaranteed bandwidth lower limit or ratio, the packet loss tolerance is the threshold for triggering the reliability protection mechanism, and the burst coefficient is the condition for determining the dynamic buffer expansion. Then, the service type is divided according to the QoS indicator parameters, where the service type includes delay-sensitive, bandwidth-intensive, reliability-sensitive and bursty services. The specific classification rules are as follows:

[0056] 1. If the delay threshold is less than or equal to the preset delay, the service request instruction is delay-sensitive;

[0057] 2. If the bandwidth quota is greater than or equal to a first predetermined percentage of the total bandwidth, the service request instruction is bandwidth-intensive;

[0058] 3. If the packet loss tolerance is less than or equal to the second predetermined percentage, the service demand instruction is reliability-sensitive;

[0059] 4. If the burst coefficient is greater than or equal to the preset burst threshold, the service demand instruction is a burst service;

[0060] Preferably, the preset delay may be 10 μs, the first preset percentage may be 30%, the second preset percentage may be 0.001%, and the preset burst threshold may be 5. Specifically, the preset delay, the first preset percentage, the second preset percentage, and the preset burst threshold may be set according to actual conditions and target requirements.

[0061] The service requirement specifies one or more of the following service types: delay-sensitive, bandwidth-intensive, reliability-sensitive, and bursty. Each condition in the above rules is evaluated independently, allowing multiple conditions to be met simultaneously. For example, a service requirement can be both delay-sensitive and bursty. For example, if the service requirement's latency threshold is 8μs (≤10μs) and the burst factor is 6 (≥5), then it is both delay-sensitive and bursty.

[0062] S12 according to the business type for the service demand instruction matching the corresponding virtual transmission channel, the virtual transmission channel is used to isolate the hardware-level resources of different types of traffic;

[0063] Step S12 includes the following:

[0064] 1. If the service request instruction belongs to a service type, the RNIC network card is matched with the corresponding virtual transmission channel according to the service type;

[0065] 2. If the service demand instruction belongs to multiple business types at the same time, the RNIC network card matches the virtual transmission channel corresponding to each business type according to the preset priority rules. The preset priority matching rules are that delay-sensitive types take precedence over reliability-sensitive types, reliability-sensitive types take precedence over bandwidth-intensive types, and bandwidth-intensive types take precedence over bursty businesses.

[0066] Here, the RNIC is a dedicated network interface card (NIC) that supports Remote Direct Memory Access (RDMA). If the service request instruction belongs to only one service type, the RNIC matches the corresponding virtual transmission channel based on that service type. If the service request instruction belongs to multiple service types, the RNIC matches the virtual transmission channel corresponding to each service type based on preset priority rules.

[0067] Specifically, the priority of a service type is defined according to its quality of service requirement (SLA): Delay-sensitive: the highest priority, requiring minimal end-to-end delay (e.g., ≤10μs); Reliability-sensitive: the second highest priority, requiring zero packet loss or an extremely low packet loss rate (e.g., ≤0.001%); Bandwidth-intensive: the medium priority, requiring high throughput (e.g., ≥10Gbps); Burst-intensive: the lowest priority, allowing dynamic adjustment and resource borrowing.

[0068] The priority of the virtual channel is configured at the hardware level (RNIC) and is used to achieve resource isolation and differentiated scheduling. The specific configuration rules are as follows: the first virtual transmission channel is the highest priority virtual channel (VL0), which is allocated to delay-sensitive services and adopts strict priority queuing; the second virtual transmission channel is the second highest priority virtual channel (VL1), which is allocated to bandwidth-intensive services and adopts weighted fair queuing; the third virtual transmission channel is a medium priority virtual channel (VL2), which is allocated to reliability-sensitive services and adopts the PFC flow control mechanism. The PFC flow control mechanism is a priority-based flow control mechanism used to prevent packet loss during network congestion. When the buffer is close to full due to traffic of a certain priority, PFC will send a "pause frame" to notify the upstream device to temporarily stop sending data of that priority, while allowing traffic of other priorities to continue to be transmitted; the fourth virtual transmission channel is the lowest priority virtual channel (VL3), which is allocated to bursty services and supports dynamic buffer expansion.

[0069] It should be noted that high-priority services (such as delay-sensitive services) will be assigned to high-priority virtual channels (such as VL0). This mapping relationship is preset to ensure that service requirements match hardware resources. The priority of the virtual channel is guaranteed to meet the service quality requirements of the corresponding service through hardware mechanisms (such as queue scheduling and preemption). For example, the strict priority queue of VL0 ensures that delay-sensitive traffic is always scheduled first. Strict priority queue means that the RNIC hardware scheduler prioritizes VL0 traffic and only processes the next priority queue (such as VL1) when the VL0 queue is empty. Hardware interrupt preemption means that if the VL0 traffic delay approaches a threshold (such as 90% of the SLA delay), the RNIC directly interrupts the transmission of the low-priority virtual channel (such as VL3) to release resources for VL0.

[0070] S13 selects the optimal transmission path for each virtual transmission channel traffic according to the preset path evaluation model;

[0071] Step S13 includes the following:

[0072] S131 obtains all available paths under the current network topology through the SDN controller and filters the candidate paths that meet the preset conditions. The preset conditions include that the delay of the current path is less than or equal to a preset multiple of the service delay threshold, and the remaining bandwidth of the current path is greater than or equal to half of the service bandwidth quota;

[0073] All available paths refer to the physical or logical transmission paths available within the current network topology that meet the basic service level agreement (SLA) requirements. The following steps determine this: 1. Dynamic discovery by the SDN controller: This uses the OpenFlow protocol to obtain network-wide link status, generate a list of available paths, and filter out paths that fail to meet basic requirements, such as latency > 100μs and remaining bandwidth < 1Gbps. The SDN controller is the core component of the software-defined network, responsible for centralized management and dynamic scheduling of network resources. 2. Statically configured path pool: This predefines several candidate paths, such as Path A, Path B, and Path C, and periodically updates their status. 3. Routing protocol generation: This dynamically calculates the shortest path or load-balancing path based on protocols such as OSPF and BGP.

[0074] Specific rules for determining candidate paths: latency ≤ 1.5 times the service latency threshold; remaining bandwidth ≥ 50% of the service bandwidth quota; and packet loss rate ≤ 2 times the service packet loss tolerance. Each virtual channel retains a maximum of three candidate paths: the primary path and two backup paths.

[0075] S132 real-time collection of evaluation parameters of each candidate path, including link delay, remaining bandwidth, packet loss rate and historical path stability;

[0076] First, build a path evaluation model, as follows:

[0077] Service type demand analysis: Latency-sensitive services prioritize paths with the lowest latency; bandwidth-intensive services prioritize paths with the highest remaining bandwidth; reliability-sensitive services prioritize paths with the lowest packet loss rate; and bursty services prioritize paths with strong dynamic adjustment capabilities. Path evaluation parameter definitions: Link delay is the end-to-end transmission delay; Remaining bandwidth is the unused bandwidth of the current link; Packet loss rate is the average packet loss rate of the link; Path stability is the historical stability score of the link.

[0078] Based on the service type requirements, a differentiated path scoring formula is designed for each virtual transmission channel, as shown below:

[0079] Latency-sensitive (VL0):

[0080]

[0081] Among them: S1 is the VL0 score corresponding to the delay-sensitive type, L is the delay, RB is the remaining bandwidth, PLR is the packet loss rate, α, β and γ are the corresponding weight coefficients respectively, α>β, γ, emphasizing low delay.

[0082] Bandwidth-intensive (VL1):

[0083]

[0084] Among them: S2 is the bandwidth-intensive corresponding VL1 score, β>α, γ, emphasizing high bandwidth.

[0085] Reliability sensitive (VL2):

[0086]

[0087] Among them: S3 is the reliability-sensitive type corresponding to the VL2 score, γ>α, β, emphasizing low packet loss.

[0088] Burst services (VL3):

[0089]

[0090] Among them: S4 is the VL3 score corresponding to bursty services, δ>α, β, emphasizing stability and flexibility.

[0091] Dynamically adjust α, β, γ, and δ based on real-time network load and service priority. For example, during peak traffic periods, the δ weight of VL3 is automatically increased.

[0092] Next, based on the path evaluation model constructed above, network probes are used to periodically measure link latency, remaining bandwidth, and packet loss rate. RNIC counters are used to obtain real-time queue depth and buffer status for virtual channels. RNIC counters are hardware counters in network cards that support Remote Direct Memory Access (RDMA) and are used to monitor and record network communication status in real time. Based on the link load and topology information from the global view of the SDN controller, parameters such as latency, remaining bandwidth, and packet loss rate are normalized to a range of 0-1. The corresponding formula is selected based on the virtual transmission channel type, and the normalized parameters are substituted into the formula to calculate the score.

[0093] S133. Calculate the score of each candidate path according to the evaluation parameters, and select the path with the highest score as the optimal transmission path.

[0094] The score calculation formula is as follows:

[0095]

[0096] Where: L is the delay, RB is the remaining bandwidth, PLR is the packet loss rate, HPS is the historical path stability, α, β, γ and δ are the corresponding weight coefficients respectively.

[0097] All available paths are sorted by score from high to low. The path with the highest score is selected as the primary path, and the suboptimal path is used as the backup. If multiple paths have the same score, the path with the highest historical stability is selected first.

[0098] Taking a financial transaction scenario as an example, the SDN controller discovers links between switches using the LLDP protocol, generates a path list, and filters out paths with delays greater than 15μs. For latency-sensitive services (α = 0.6, β = 0.2, γ = 0.1, δ = 0.1), path A scores 9.2, while path B scores 8.5. Path A is selected.

[0099] Through the above steps, the path evaluation model can dynamically select the optimal path based on the needs of different business types. Combined with real-time data collection, dynamic weight adjustment, and hardware-level preemption mechanisms, it can ensure that delay-sensitive traffic always chooses the fastest path, bandwidth-intensive traffic maximizes occupied spatial bandwidth, reliability-sensitive traffic avoids high-packet-loss links, and bursty services respond to network fluctuations through multi-path load balancing.

[0100] S14. Use a prediction model to analyze the congestion risk of the optimal transmission path;

[0101] Step S14 includes the following:

[0102] S141. Obtain the data source of the optimal transmission path from the SDN controller and RNIC;

[0103] S142. Generate time series feature data from the data source according to a preset time window, and input the time series feature data into the trained LSTM prediction model;

[0104] S143. Output the congestion probability and congestion risk level of the optimal transmission path within a preset time.

[0105] The training process of the LSTM prediction model is as follows:

[0106] First, obtain the data source, which includes real-time network status data, historical traffic data, service type parameters, and path stability indicator data. Real-time network status data includes link delay, remaining bandwidth, packet loss rate, queue depth, and buffer occupancy rate. Historical traffic data is the traffic pattern over a period of time, such as traffic periodicity and burst characteristics. Service type parameters include the delay threshold, bandwidth quota, packet loss tolerance, and burst coefficient in the SLA label. Path stability indicator data includes the path's packet loss rate fluctuation, delay standard deviation, and historical interruption count.

[0107] Normalize parameters of different dimensions in the data source to a range of 0-1: Aggregate data by time window to generate time series feature data; and filter out noisy data caused by instantaneous jitter. Specifically, the input features of the LSTM prediction model (Long Short-Term Memory) include time series features, path features, and service features. Time series features include link delay, bandwidth utilization, and packet loss rate for the past N time slices; path features include historical stability scores, current queue depth, and burst coefficient; and service features include delay thresholds and bandwidth quotas in SLA labels. Output targets include congestion probability and congestion level. Congestion probability is the probability (0-1) that the path will be congested within the next T time slices (e.g., 10ms); congestion levels include low risk, medium risk, and high risk.

[0108] The real-time inference steps are as follows:

[0109] Current path status data is obtained from the SDN controller and RNIC, and a time series is generated by time window, such as data from the past 10 time slices, which is then fed into a trained prediction model. The input features are passed through the prediction model, and the output is the probability of future congestion.

[0110] For example, if the congestion probability of path A is 85% and that of path B is 20%, path A is considered high risk. The risk classification is as follows: Low risk (probability < 30%): No adjustments required, continuous monitoring; Medium risk (30% ≤ probability < 70%): Alert and prepare backup paths; High risk (probability ≥ 70%): Immediate path switching.

[0111] S15. Dynamically adjust traffic transmission paths based on congestion risk analysis results, and ensure the transmission quality of the highest priority services through hardware-level scheduling strategies.

[0112] Specifically, dynamically adjusting traffic transmission paths based on congestion risk analysis results includes:

[0113] 1. When the congestion probability is greater than or equal to a third preset percentage, the transmission path is switched from the optimal transmission path to an alternative path, where the alternative path is the available path with the second highest path score and a congestion probability less than the third preset percentage;

[0114] 2. When the congestion probability is greater than or equal to a fourth preset percentage and less than a third preset percentage, half of the traffic in the optimal transmission path is migrated to the alternative path.

[0115] Preferably, the third preset percentage may be 70%, and the fourth preset percentage may be 30%.The third preset percentage and the fourth preset percentage may be set according to actual circumstances and desired purposes.

[0116] Specifically, the alternative path is selected based on the path score. The path score calculation method is described in the above steps and is not repeated here. When the congestion probability is greater than or equal to 30% and less than 70%, a portion of the traffic on the optimal path is migrated to the alternative path. In this case, the portion of traffic is set to 50%.

[0117] The transmission quality of the highest priority services is guaranteed through hardware-level scheduling strategies, including:

[0118] Specifically, the hardware-level scheduling strategy is an RNIC function call, which includes:

[0119] 1. When the queue depth of the first virtual transmission channel matching the delay-sensitive service is greater than or equal to a fifth preset percentage of the total capacity, suspend traffic transmission on all virtual transmission channels except the first virtual transmission channel through the RNIC hardware scheduler;

[0120] 2. Cache the suspended traffic to an independent backup queue, and resume traffic transmission on other virtual transmission channels when the queue depth of the first virtual transmission channel is less than 50%.

[0121] Preferably, the fifth preset percentage may be 80%, and the fifth preset percentage may be set according to actual circumstances and desired purposes.

[0122] In addition, when the buffer for reliability-sensitive traffic (VL2) is close to full, the RNIC sends a PFC frame to pause the peer's transmission. During the pause, the traffic is cached in an independent queue to avoid packet loss; bursty services (VL3) expand the cache from the shared buffer pool on demand.

[0123] For example, when delay-sensitive services (VL0) and bursty services (VL3) are transmitted simultaneously, VL0 traffic is immediately sent through the strict priority queue, with a stable latency of 5 seconds. VL3 traffic is transmitted when the VL0 queue is idle. If VL0 traffic bursts, the RNIC interrupts VL3 traffic and caches it in the backup queue. The latency of VL0 is always ≤10us, and the throughput of VL3 may decrease, but it will not affect high-priority services.

[0124] Finally, real-time monitoring and feedback optimization are performed. Monitoring metrics include the real-time performance of each path (latency, bandwidth, and packet loss rate); virtual channel resource utilization (queue depth and buffer utilization); and service SLA compliance (e.g., latency compliance ≥ 99.9%). The feedback mechanism compares actual congestion events with predicted results to adjust LSTM model parameters; dynamically updates path score weights based on historical stability (e.g., reducing the α value for frequently congested paths); and hardware policy tuning: adjusting the preemption threshold based on interrupt frequency, such as from 90% to 85%.

[0125] Through the aforementioned dynamic congestion risk assessment, path switching, hardware-level scheduling, and monitoring feedback, traffic can be preemptively migrated based on predictive models to minimize service interruptions. RNIC hardware acceleration enables microsecond-level resource preemption, achieving an optimal balance between service quality and resource efficiency in complex mixed service environments.

[0126] Example 2

[0127] An embodiment of a network scheduling system for hybrid service transmission based on RDMA in the present invention includes the following steps:

[0128] A service requirement instruction receiving unit, configured to receive a service requirement instruction to be transmitted and determine a corresponding service type according to a quality of service parameter in the service requirement instruction;

[0129] A virtual transmission channel matching unit, configured to match a corresponding virtual transmission channel to the service demand instruction according to the service type, wherein the virtual transmission channel is used to isolate hardware-level resources for different types of traffic;

[0130] An optimal transmission path selection unit, configured to select an optimal transmission path for the traffic of each virtual transmission channel according to a preset path evaluation model;

[0131] A congestion risk prediction unit, configured to analyze the congestion risk of an optimal transmission path using a prediction model;

[0132] The traffic transmission path adjustment unit is used to dynamically adjust the traffic transmission path according to the congestion risk analysis results, and to ensure the transmission quality of the highest priority services through hardware-level scheduling strategies.

[0133] The specific functions and uses of the units in this embodiment are similar to the steps in the aforementioned embodiment 1 and are not described in detail here.

[0134] It is understandable that those skilled in the art can, under the guidance of the above embodiments, combine various implementation methods in the above embodiments to obtain technical solutions of multiple implementation methods.

[0135] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A network scheduling method for hybrid service transmission based on RDMA, characterized in that: include: receiving a service requirement instruction to be transmitted, and determining a corresponding service type according to a quality of service parameter in the service requirement instruction; Matching a corresponding virtual transmission channel to the service demand instruction according to the business type, wherein the virtual transmission channel is used to isolate hardware-level resources of different types of traffic; Select the optimal transmission path for the traffic of each virtual transmission channel according to the preset path evaluation model; Analyzing the congestion risk of the optimal transmission path using a prediction model; Dynamically adjust traffic transmission paths based on congestion risk analysis results, and ensure the transmission quality of the highest priority services through hardware-level scheduling strategies.

2. The network scheduling method for hybrid service transmission based on RDMA according to claim 1, characterized in that: The receiving of a service requirement instruction to be transmitted and determining a corresponding service type according to a quality of service parameter in the service requirement instruction includes: If the delay threshold is less than or equal to the preset threshold, the service demand instruction is delay-sensitive; If the bandwidth quota is greater than or equal to a first predetermined percentage of the total bandwidth, the service demand instruction is bandwidth-intensive; If the packet loss tolerance is less than or equal to the second preset percentage, the service demand instruction is reliability-sensitive; If the burst coefficient is greater than or equal to the preset burst threshold, the service demand instruction is a burst service; The service demand instruction belongs to one or more service types selected from delay-sensitive, bandwidth-intensive, reliability-sensitive and bursty services.

3. The network scheduling method for hybrid service transmission based on RDMA according to claim 2, characterized in that: The matching of the service demand instruction to a corresponding virtual transmission channel according to the business type, wherein the virtual transmission channel is used to isolate hardware-level resources for different types of traffic, includes: If the service demand instruction belongs to a service type, matching a corresponding virtual transmission channel according to the service type through the RNIC network card; If the service demand instruction belongs to multiple business types at the same time, the RNIC network card will match the virtual transmission channel corresponding to each business type according to the preset priority rules. The preset priority matching rules are that delay-sensitive type takes precedence over reliability-sensitive type, reliability-sensitive type takes precedence over bandwidth-intensive type, and bandwidth-intensive type takes precedence over bursty business.

4. The network scheduling method for hybrid service transmission based on RDMA according to claim 3, characterized in that: The method further includes matching a corresponding virtual transmission channel for the service demand instruction according to the business type, wherein the virtual transmission channel is used to isolate hardware-level resources for different types of traffic. The delay-sensitive service is matched with a first virtual transmission channel, and the first virtual transmission channel adopts a strict priority queue; The reliability-sensitive service is matched to a second virtual transmission channel, and the second virtual transmission channel is a weighted fair queue; The bandwidth-intensive service is matched with the third virtual transmission channel, and the third virtual transmission channel adopts the PFC flow control mechanism; The bursty service matches the fourth virtual transmission channel, and the fourth virtual transmission channel supports dynamic buffer expansion.

5. The network scheduling method for hybrid service transmission based on RDMA according to claim 1, characterized in that: The selecting the optimal transmission path for the traffic of each virtual transmission channel according to the preset path evaluation model includes: The SDN controller obtains all available paths in the current network topology and selects the candidate paths that meet the preset conditions. The preset conditions include that the delay of the current path is less than or equal to a preset multiple of the service delay threshold, and the remaining bandwidth of the current path is greater than or equal to half of the service bandwidth quota; Collect evaluation parameters of each candidate path in real time, including link delay, remaining bandwidth, packet loss rate and historical path stability; The scores of the candidate paths are calculated according to the evaluation parameters, and the path with the highest score is selected as the optimal transmission path.

6. The network scheduling method for hybrid service transmission based on RDMA according to claim 5, characterized in that: Calculating the scores of the candidate paths according to the evaluation parameters and selecting the path with the highest score as the optimal transmission path includes: Where: L is the delay, RB is the remaining bandwidth, PLR is the packet loss rate, HPS is the historical path stability, α, β, γ and δ are the corresponding weight coefficients respectively.

7. The network scheduling method for hybrid service transmission based on RDMA according to claim 1, characterized in that: The adopting a prediction model to analyze the congestion risk of the optimal transmission path includes: Obtaining the data source of the optimal transmission path from the SDN controller and the RNIC; The data source generates time series feature data according to a preset time window, and the time series feature data is input into the trained LSTM prediction model; Output the congestion probability and congestion risk level of the optimal transmission path within a preset time.

8. The network scheduling method for hybrid service transmission based on RDMA according to claim 1, characterized in that: The method of dynamically adjusting traffic transmission paths based on congestion risk analysis results and ensuring the transmission quality of the highest priority services through hardware-level scheduling strategies includes: When the congestion probability is greater than or equal to a third preset percentage, switching the transmission path from the optimal transmission path to an alternative path, where the alternative path is an available path with a second-ranked path score and a congestion probability less than a third preset percentage; When the congestion probability is greater than or equal to a fourth preset percentage and less than a third preset percentage, half of the traffic in the optimal transmission path is migrated to the alternative path.

9. The network scheduling method for hybrid service transmission based on RDMA according to claim 8, characterized in that: The method of dynamically adjusting traffic transmission paths based on congestion risk analysis results and ensuring the transmission quality of the highest priority services through hardware-level scheduling strategies also includes: When the queue depth of the first virtual transmission channel matching the delay-sensitive service is greater than or equal to a fifth preset percentage of the total capacity, suspending traffic transmission of other virtual transmission channels except the first virtual transmission channel through the RNIC hardware scheduler; The suspended traffic is cached in an independent backup queue, and traffic transmission on other virtual transmission channels is resumed when the queue depth of the first virtual transmission channel is less than 50%.

10. A network scheduling system for hybrid service transmission based on RDMA, characterized in that: A network scheduling method for hybrid service transmission based on RDMA according to any one of claims 1 to 9, comprising: A service requirement instruction receiving unit, configured to receive a service requirement instruction to be transmitted and determine a corresponding service type according to a quality of service parameter in the service requirement instruction; A virtual transmission channel matching unit, configured to match a corresponding virtual transmission channel to the service demand instruction according to the service type, wherein the virtual transmission channel is used to isolate hardware-level resources for different types of traffic; An optimal transmission path selection unit, configured to select an optimal transmission path for the traffic of each virtual transmission channel according to a preset path evaluation model; a congestion risk prediction unit, configured to analyze the congestion risk of the optimal transmission path using a prediction model; The traffic transmission path adjustment unit is used to dynamically adjust the traffic transmission path according to the congestion risk analysis results, and to ensure the transmission quality of the highest priority services through hardware-level scheduling strategies.

Images