Fair congestion control method and device with flow awareness in SDN data center network
The flow-aware fair congestion control method optimizes SDN data center networks by using a fine-grained scheduling algorithm and coupled BBR congestion control to enhance throughput and reduce packet loss through fair bandwidth allocation.
Patent Information
- Application Number
- CN202311727381.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-14
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2043-12-14
AI Technical Summary
Network congestion in SDN data center networks leads to increased packet loss, end-to-end delay, and reduced resource utilization due to limited buffer capacity, bandwidth, and processing capacity, exacerbated by uneven traffic load distribution and inefficient load balancing.
A flow-aware fair congestion control method using a fine-grained data scheduling algorithm, coupled BBR congestion control, and bandwidth allocation to adjust sending rates based on network conditions, distinguishing between large and small flows to optimize network performance.
The method enhances network throughput, reduces packet loss, and maintains stable performance by ensuring fair bandwidth allocation and reducing queuing times, particularly in burst traffic conditions.
Smart Images

Figure CN117714381B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of communication technology, and in particular to a flow-aware fair congestion control method, device and storage medium in an SDN data center network. Background Art
[0002] In the communication process of data centers, network congestion is inevitable, which is mainly due to the limited buffer capacity of network equipment, the limited bandwidth capacity of data links, and the limited processing capacity of network nodes. Network congestion will lead to increased packet loss rate, increased end-to-end delay and reduced resource utilization, and even cause huge losses to large Internet companies.
[0003] Software Defined Networking (SDN) is a new network paradigm used to manage data flow transmission in computer networks. It performs unified network resource scheduling and implements flexible bandwidth allocation through a logically centralized control plane, thereby improving network resource utilization. By utilizing the technical advantages of SDN centralized control and flexible scheduling, deploying an efficient and secure multi-path transmission mechanism in the data center network will further improve data transmission performance and comprehensively improve the network's service quality. However, the communication mode of the decentralized and aggregated workflow of the data center switch will inevitably lead to congestion problems. In addition, there are usually multiple paths for data transmission between the source and destination terminals in the data center. If load balancing is not performed properly, it is easy to cause uneven traffic load distribution. Summary of the invention
[0004] The present invention proposes a flow-aware fair congestion control method, device and storage medium in an SDN data center network, which can at least solve one of the technical problems in the background technology.
[0005] To achieve the above object, the present invention adopts the following technical solutions:
[0006] A fair congestion control method with flow awareness in an SDN data center network comprises the following steps:
[0007] S1. Design a fine-grained data scheduling algorithm;
[0008] S2, performing congestion control of data flow based on data packet scheduling implemented by the data scheduling algorithm;
[0009] S3, adjusting the sending rates of different data flows based on the current network environment, i.e., calculating the sub-flow rate weights and implementing bandwidth allocation based on the congestion control algorithm in step S2;
[0010] S4, finally executing the optimal congestion control algorithm to make the network performance reach the optimal performance.
[0011] On the other hand, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.
[0012] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.
[0013] It can be seen from the above technical solutions that the fair congestion control method and device with flow perception in the SDN data center network of the present invention fully considers the attributes of different services when designing the scheduling algorithm, distinguishes between large flows and small flows, and avoids the problem of disordered data packets cached at the receiving end. In addition, the present invention designs a fair congestion control algorithm with flow perception based on the BBR congestion control algorithm, which ensures fairness to traditional single-path flows while satisfying the improvement of network performance. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 Schematic diagram of a BBR congestion control model based on flow perception according to an embodiment of the present invention;
[0015] Figure 2 This is a schematic diagram of the fat tree topology;
[0016] Figure 3 The experimental results of the first group of experiments shown are based on data throughput as the performance indicator;
[0017] Figure 4 What is shown are the experimental results with packet loss rate as the performance indicator;
[0018] Figure 5 It is a comparison chart of packet loss rate between DCTCP and the network of the present invention. DETAILED DESCRIPTION
[0019] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.
[0020] The embodiment of the present invention proposes a new scheduling algorithm based on the coupled BBR congestion control. This is because the existing predictive scheduling algorithms are based on the growth law of the window. This type of predictive scheduling algorithm will no longer be applicable after the underlying congestion control algorithm becomes the BBR congestion control algorithm. Therefore, a new predictive scheduling algorithm needs to be proposed.
[0021] The congestion control algorithms commonly used in SDN are usually packet loss-based or delay-based algorithms. These algorithms avoid congestion by triggering window reduction. In a multipath transmission mechanism, this behavior will trigger load migration, resulting in a decrease in overall transmission performance. Google proposed a hybrid congestion control algorithm BBR, which has received a lot of attention since its release and has been applied to a variety of network architectures to improve network transmission. BBR adjusts the sending rate by measuring the network bottleneck bandwidth and minimum delay in real time. The maximum transmission rate in the last 10 round-trip delays (RTT) is regarded as the bottleneck bandwidth (BW), and the minimum delay measured in the past 10 seconds is regarded as RTprop. Then, the pacing gain and window gain are used to control its sending behavior. The sending rate (pacing_rate / sending_rate) and congestion window (CWND) are:
[0022] pacing_rate(sending_rate)=pacing_gain×BW (6)
[0023] CWND=cwnd_gain×BW×RTprop (7)
[0024] Where pacing_gain and cwnd_gain are pacing gain and window gain respectively.
[0025] Due to the excellent performance of BBR, it is extended to the multi-path congestion control scenario in SDN, and a BBR congestion control scheme based on flow perception is proposed, such as Figure 1 As shown, the following steps are included:
[0026] S1. Design a fine-grained data scheduling algorithm;
[0027] S2, performing congestion control of data flow based on data packet scheduling implemented by the data scheduling algorithm;
[0028] S3, adjusting the sending rates of different data flows based on the current network environment, i.e., based on the calculation of sub-flow rate weights and bandwidth allocation in the congestion control algorithm in step S2;
[0029] S4, finally executing the optimal congestion control algorithm to make the network performance reach the optimal performance.
[0030] The following are respectively explained:
[0031] The embodiment of the present invention designs a fine-grained data scheduling algorithm S1 before executing the coupled BBR algorithm. The specific steps include:
[0032] S1.1, obtain the BW and RTT of each subflow in the initial link respectively.
[0033] S1.2, for each data packet in the queue, a path with a smaller link load is selected as the route according to the load conditions of each link, and a suitable sub-flow is selected for transmission in FIFO order according to its arrival time.
[0034] S1.3,In order to avoid long queuing time for data packet transmission, when the sender detects a packet loss event, it will update the packet loss rate and re-predict the time to arrive at the receiver.
[0035] S1.4, re-schedule data packets
[0036] S1.4.1, according to the characteristics of large and small data flows in the data center network, the change in bytes transmitted is periodically detected to distinguish between large and small flows, and data packet scheduling is performed.
[0037] S1.4.2, select the most suitable large flow for transfer. Specifically, when scheduling connection-level data packets, estimate the arrival time of the data packets in each sub-flow, and select the sub-flow with the earliest arrival time each time.
[0038] S2: Coupled BBR congestion control, the specific steps include:
[0039] S2.1,When designing, we need to ensure fairness of bottleneck links while improving performance. We cannot simply superimpose BBR. Specifically, we further analyze the data flow sending rate model. Assume that there is a bottleneck link with a bandwidth of C and n different RTT flows are passing through it. Let flow i i∈[1,n] represents flow i , d i (t) indicates flow i The transfer rate, under ideal conditions, is d1+d2+…+d n =C.I i (t) indicates flow i The amount of in-flight data, i.e. the upper bound of the bottleneck link, is calculated by formula (3):
[0040]
[0041] Combining formulas (1) and (3), flow i In the first upward detection phase, a gain of 1.25 times can be obtained, so the maximum transfer rate at time t is:
[0042]
[0043] Where d i (t) indicates flow i The transfer rate, For flow iThe detection period of BBR flow is 8RTprop, then flow i In the new round the bandwidth estimates are updated to:
[0044]
[0045] From formula (5), we can see that the actual gain of the data flow on the same bottleneck link is However, the RTTs of data streams are not exactly the same, so it is difficult to achieve fair sharing of their initial bandwidth.
[0046] S2.2, fairness optimization between BBR data flows. The specific steps include:
[0047] S2.2.1, according to RTT, the link utilization ω at the bottleneck can be quantified, expressed as flow i The percentage of RTT and maximum link delay Tmax.
[0048] S2.2.2, determine the congested link. When the bandwidth utilization of the link reaches 90%, it is determined that the link is congested and the data flow in it needs to be rerouted to alleviate the congestion of the link.
[0049] S2.2.3, division of bottleneck set. If two or more sub-streams simultaneously experience a linear increase in RTT, it is preliminarily determined that these sub-streams share the same bottleneck. The sub-stream rates within the same bottleneck are coupled, and the pacing gains of different sub-streams are adjusted according to the link status to balance the sending rates between data streams. Step S2.2.3 specifically includes:
[0050] S2.2.3.1,According to the characteristics of large and small data flows in the data center network, small flows are more sensitive to latency, so generally only large flows are restricted.
[0051] S2.2.3.2, Calculation of sub-stream rate weights. The currently measured bandwidth is proportionally allocated to the rate of each sub-stream. The parameter α is calculated as follows:
[0052]
[0053] S2.2.3.3, Balanced bandwidth. When multiple sub-flows compete with each other on the bottleneck link, the bandwidth obtained by each sub-flow is proportional to the size of the buffer occupied by it in the link queue. The parameter α is used to regulate the competition for bandwidth among the sub-flows. Combined with formula 1, the sending rate is updated as:
[0054] pacing_rate(sending_rate)=α×pacing_gain×BW (12)
[0055] S3, adjusting the sending rates of different data streams based on the current network environment.
[0056] S4, executing the optimal congestion control algorithm to make the network performance reach the optimal performance.
[0057] The following examples illustrate the advantages of the embodiments of the present invention:
[0058] First, the proposed optimization scheme is evaluated based on the NS3 simulation tool. The simulation code is based on the source code provided by Googlegroup. Then the optimization scheme is implemented in the Linux kernel and tested. Figure 2 The fat tree topology shown in Figure 2. In the comparison of the optimization schemes, we used DCTCP, a congestion control algorithm commonly used in SDN networks, to compare with the optimization schemes.
[0059] To avoid network congestion, the congestion control algorithm should be able to overcome the degradation of the overall network throughput performance. Figure 3 The first group of experiments shown in the figure uses data throughput as the performance indicator. The experiment sets up 23 hosts (host1 to host23) to transmit more than 2GB of data to the same host (host24), and the link bandwidth is 1GB. By monitoring and sampling the data throughput from host 24 to its edge switch, the throughput performance of DCTCP and the optimization scheme is compared.
[0060] Figure 3 The fluctuation of the data throughput transmitted in the first 200 seconds of the entire transmission process is shown, among which the throughput of the optimized solution is significantly higher than that of DCTCP. The throughput of DCTCP fluctuates frequently and violently, and its average throughput is around 757Mbps. This shows that when network congestion occurs, the bandwidth resources of DCTCP are not fully occupied and cannot effectively eliminate network congestion. In the optimized solution, the throughput performance drops slightly and then remains stable at around 806Mbps. Compared with DCTCP, its throughput has increased by 6.47%, which means that even if network congestion occurs, the performance of the bandwidth throughput of the optimized solution can stably maintain high quality.
[0061] Flow completion time (FCT) is an important indicator of an application and will directly affect the performance quality of the application. Figure 3 The results are shown in the experiment with data completion time as the performance indicator. The experimental design is that the client sends different sizes of streams to the server concurrently, including small streams (<100KB), medium streams (100KB~10MB) and large streams (>10MB), and measures the average stream completion time of DCTCP and the optimization scheme under different concurrent traffic sizes.
[0062] from Figure 4As can be seen in the figure, for small flows (<1MB), the completion time performance of DCTCP and the optimized solution in a congested environment is not much different, and the optimized solution improves by 19.3%. For medium flows (100KB~10MB), especially when the data flow scale is greater than 1MB, the FCT of the optimized solution is about 22.5% lower than that of DCTCP. However, for large flows (>10MB), the increase in FCT of the optimized solution is greater than that of DCTCP, but the FCT of the optimized solution is still lower than that of DCTCP. Overall, compared with DCTCP, the optimized solution has a lower completion time and provides better bandwidth for data flows.
[0063] Figure 4 The experimental results with packet loss rate as the performance indicator are displayed, that is, the algorithm's flow completion time. The experiment introduces burst traffic in a stable background flow environment to cause network congestion. During the 10 to 60 seconds of the experiment, iperf is used to generate background flows in the network, and each host sends a data flow to the specified destination host.
[0064] Figure 5 This is a comparison chart of network packet loss rates. Figure 5 It can be seen that when there are less than 2 burst flows, the packet loss rates of DCTCP and the optimization scheme are 0.78% and 0.85% respectively. As the number of burst flows increases, the anti-packet loss advantage of the optimization scheme becomes more obvious. The optimization scheme reduces the overall packet loss rate by 16.3% compared with DCTCP. Through the above experiments, it can be observed that the optimization scheme can effectively reduce the packet loss rate, which shows that the optimization scheme is stable in dealing with burst flow conditions and can present better network performance.
[0065] In summary, the present invention fully considers the attributes of different services when designing the scheduling algorithm, distinguishes between large flows and small flows, and avoids the problem of disordered data packets cached at the receiving end.
[0066] On the other hand, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.
[0067] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.
[0068] In another embodiment provided in the present application, a computer program product including instructions is also provided, which, when executed on a computer, enables the computer to execute the fair congestion control method with flow awareness in an SDN data center network in any of the above embodiments.
[0069] It is understandable that the system provided by the embodiment of the present invention corresponds to the method provided by the embodiment of the present invention, and the explanation, examples and beneficial effects of the relevant contents can refer to the corresponding parts in the above method.
[0070] The embodiment of the present application also provides an electronic device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus.
[0071] Memory, used to store computer programs;
[0072] The processor is used to implement the fair congestion control method with flow perception in the SDN data center network when executing the program stored in the memory.
[0073] The communication bus mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc.
[0074] The communication interface is used for communication between the above electronic device and other devices.
[0075] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.
[0076] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, and discrete hardware components.
[0077] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive Solid State Disk (SSD)), etc.
[0078] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
[0079] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.
[0080] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A flow-aware fair congestion control method in an SDN data center network, characterized in that: The following steps are included: S1. Design a fine-grained data scheduling algorithm; S2, performing congestion control of data flow based on data packet scheduling implemented by the data scheduling algorithm; S3, adjusting the sending rates of different data flows based on the current network environment, i.e., based on the calculation of sub-flow rate weights and bandwidth allocation in the congestion control algorithm in step S2; S4, finally executing the optimal congestion control algorithm to achieve the optimal network performance; The S2, based on the data packet scheduling implemented by the data scheduling algorithm, performs congestion control on the data flow, specifically including: S2.1, through the data flow sending rate model, further analysis, assume that there is a bottleneck link with a bandwidth of C, and n different RTT flows are passing through; Represents n different RTT streams, d i (t) means flow i The transfer rate, under ideal conditions, d 1+ d 2+…+ d n =C; I i (t) means flow i The amount of in-flight data, i.e. the upper bound of the bottleneck link, is calculated by formula (3): (3) flow i In the first upward detection phase, a gain of 1.25 times can be obtained, so the maximum transfer rate at time t is: (4) In the formula d (t) means flowi The transfer rate, for flowi RTprop; the detection period of BBR flow is 8RTprop, then flowi In the new round the bandwidth estimates are updated to: (5) From formula (5), we can see that the actual gain of the data flow on the same bottleneck link is , and the RTT of data streams will not be exactly the same, so it is difficult to achieve fair sharing of its initial bandwidth; BBR regards the maximum delivery rate in the last 10 RTTs as BtIBw, and the minimum delay measured in the past 10 seconds as RTprop; S2.2, fairness optimization between BBR data flows, specifically including: S2.2.
1. Calculate the link utilization ω at the RTT quantization bottleneck. The link utilization ω is expressed as flow i The percentage of RTT and maximum link delay Tmax; S2.2.2, determine the congested link; when the bandwidth utilization of the link reaches 90%, it is determined that the link is congested and the data flow needs to be rerouted to alleviate the congestion of the link; S2.2.3, division of bottleneck set; if two or more sub-flows simultaneously show the phenomenon of linear increase in RTT, it is preliminarily determined that these sub-flows share the same bottleneck; the sub-flow rates within the same bottleneck are coupled, and the pacing gains of different sub-flows are adjusted according to the link status to balance the sending rates between data flows; Step S2.2.3 specifically includes: S2.2.3.
1. Based on the characteristics of large and small data flows in the data center network, small flows are more sensitive to latency, so only large flows are restricted; S2.2.3.
2. Calculation of sub-stream rate weights: The currently measured bandwidth is proportionally allocated to the rate of each sub-stream, parameter The calculation formula is as follows: (6) S2.2.3.3, balanced bandwidth; when multiple sub-flows compete with each other on the bottleneck link, the bandwidth obtained by each sub-flow is proportional to the size of the buffer it occupies in the link queue, using the parameter To regulate the competition of each sub-flow for bandwidth, the sending rate is updated to: (7)。 2. According to claim 1, the fair congestion control method with flow awareness in the SDN data center network is characterized by: S1, design a fine-grained data scheduling algorithm, including, S1.
1. Obtain the bandwidth of each sub-flow in the initial link separately and RTT; S1.2, for each data packet in the queue, select the path with the smaller link load as the route according to the load of each link, and select the appropriate sub-flow for transmission in FIFO order according to its arrival time; S1.3, when the sender detects a packet loss event, it will update the packet loss rate and re-predict the time to reach the receiver; S1.4, re-schedule the data packets.
3. The fair congestion control method with flow awareness in the SDN data center network according to claim 2 is characterized in that: S1.4, re-scheduling data packets, specifically includes: S1.4.
1. Based on the characteristics of large and small data flows in the data center network, periodically detect the byte changes in transmission to distinguish large and small flows and perform data packet scheduling; S1.4.
2. Select the most suitable large flow for transfer; specifically, when scheduling connection-level data packets, estimate the arrival time of the data packets in each sub-flow, and select the sub-flow with the earliest arrival time each time.
4. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 3.
5. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 3.
