Abstract
To solve the problem of jitter and low network throughput caused by the impact of background flows on IQ traffic in mobile fronthaul network, this paper proposed a new scheduling model for background flows, named hierarchical crossover traffic scheduling mechanism based on time-aware shaper (HC-TAS) by improving the traditional counterpart. Then, in this new model, we designed an inbound scheduling algorithm based on frame length matching and an outbound scheduling algorithm based on queue status, making sure that smaller data frames will not be blocked by large data frames. This greatly improves the utilization of timeslots in the scheduling process and reduces the jitter impact of background flows. To verify its performance, we conducted experiments in a simulated fronthaul network conforming to IEEE 802.1CM. The experimental results show that, under the condition that the jitter is guaranteed to be zero, compared with two mainstream scheduling schemes, Comb-FITting and TAS + Preemption, our proposed scheme can achieve lower maximum end-to-end delay and higher link utilization. The proposed HC-TAS meets the requirements of low jitter and high bandwidth utilization in 5G fronthaul network, and the research results provide a technical basis for the application and development of general time-sensitive networks as well.
1. Introduction
In time-sensitive networks (TSN), traffic scheduling is one of the key technologies. At present, the TSN Working Group has completed the standardization of a series of traffic scheduling mechanisms, including IEEE 802.1Qav (credit value-based shaping), IEEE 802.1Qbv (gated scheduling), IEEE 802.3br and IEEE 802.1Qbu (frame preemption), IEEE 802.1Qch (cyclic queue forwarding), and IEEE 802.1Qcr (asynchronous shaping). Among them, the time-aware shaper (TAS) is the most promising. It adopts the gate control entries scheduling method, which is similar to the time division multiple access technology [1], and causes a very low jitter.
The traditional mobile fronthaul network is mainly responsible for the communication between remote radio units (RRUs) and building baseband units (BBUs). The cloud radio access network is used to virtualize the functions of BBUs so that they can transmit In-phase and Quadrature modulation streams (IQ streams) to multiple RRUs. BBU is divided into a distributed unit (DU) and a centralized unit (CU). From the RRU to the core network, it is also divided into fronthaul transmission, mid-haul transmission, and backhaul transmission. In 5G communication service [2], RRU and DU are connected through a fronthaul network. This architecture realizes the on-demand configuration of mobile fronthaul network resources through flexible and configurable CU/DU function separation, making it better adapt to personalized requirements of different scenarios. But at the same time, it puts forward strict delay requirements in such a complex traffic environment. The current cellular-based network deployment cannot meet the ultra-low delay, low jitter, and high-capacity requirements brought by the BBU division [3–5].
In this paper, we propose a hierarchical crossover time aware shaper (HC-TAS) traffic scheduling mechanism for 5G mobile fronthaul networks to achieve low latency, low jitter, and high link utilization. The main contributions are summarized as follows:(i)On the basis of the traditional TAS, we add a cross-two-level queue scheduling for background flows to ensure normal transmission of different types of traffic.(ii)An inbound scheduling algorithm and outbound scheduling algorithms are designed to effectively improve slot utilization and reduce the average latency of overall traffic.
2. Related Works
The TSN scheduling mechanisms are roughly divided into two categories, namely, synchronous traffic scheduling and asynchronous traffic scheduling [6–16]. At present, synchronous traffic scheduling mainly focuses on gate control lists (GCLs) and TAS-based joint routing. Asynchronous traffic scheduling includes credit-based shaper (CBS), asynchronous traffic shaper (ATS), and frame preemption scheduling.
2.1. Synchronous Traffic Scheduling
(1)TAS: TAS-based traffic scheduling is based on IEEE 802.1Qbv. New researches on TAS focus on timeslot calculation and allocation of control data traffic (CDT). Through reasonable modeling and algorithm design, it can ensure the boundedness of CDT delay and reduce the scheduling delay and jitter. For example, Chang et al. [17] abstracted timeslot allocation as a combinatorial optimization problem and calculated slot allocation through satisfiability modulo calculation of static scheduling. Similar to other optimization algorithms, such as integer linear programing and genetic algorithms, it is of large complexity and small scalability and difficult to adapt to medium and large-size networks.
The introduction of a guard band in TAS leads to low bandwidth utilization. To handle that, Zhao et al. [18] mapped the slot allocation problem to no-wait job scheduling and proposed a tabu search algorithm and a schedule compression technology to reduce the guard bandwidth. Chitimalla et al. [19] proposed a time slot allocation algorithm called Comb-FITting (C-FIT) for low-jitter fronthaul network scheduling. Kim et al. [20] proposed a TAS-adaptive bandwidth-sharing mechanism to improve link utilization. However, these researches have not substantially solved the problem of low bandwidth utilization caused by guard bands.(2)Joint route scheduling: By expanding TAS, Li et al. [21] proposed a joint routing method to optimize scheduling. It has higher network traffic schedulability and greatly reduces link load as well as end-to-end average delay of flows. Based on Li’s work, Bush [22] introduced hops to further reduce the average delay. Krolikowski et al. [23] optimized the routing path of AVB flow using TAS to schedule CDT flow. However, these schemes also have large guard bands, so the path selection mechanism still has a great negative impact on scheduling delays.
2.2. Asynchronous Traffic Scheduling
(1)CBS: CBS schedules traffic by setting a credit value. Using network calculus, Yang et al. [24] deduced the waiting time of service according to the upper limit of CBS’s credit value to calculate the bounded delay. Feng et al. [25] designed a specific shaping mechanism based on flow QoS by combining the strict priority algorithm and the CBS algorithm, in which the CBS algorithm is responsible for shaping ordinary flows. Maile et al. [26] proposed a scheme by combining CBS with TAS, which effectively meets the QoS requirements of industrial automation scenarios. Though these scheduling schemes can effectively reduce the delay, there exist jitters of high-priority streams caused by the continuous transmission of low-priority streams.(2)ATS: ATS defines the priority of flows based on the urgency of the event. It uses rate control strategies for parallel scheduling of high-priority flows to improve network bandwidth utilization. Zhang et al. [27] proposed a deterministic transmittable time-based asynchronous scheduler based on an urgency-based scheduler algorithm. The high-priority flow is scheduled with a rate control strategy, which improves bandwidth utilization and reduces jitter. However, in the case of multirate traffic, its scheduling complexity is greatly increased.(3)Frame preemption: Frame preemption is to prevent the continuous transmission of low-priority traffic from hindering the timely transmission of high-priority traffic. However, it is not able to prevent the influence of the background flow and eventually causes greater jitter [28]. At present, frame preemption-based scheduling is combined with other scheduling algorithms. For example, Bello et al. [29] proposed to combine the TAS mechanism with frame preemption to reduce the delay of high-priority flows and improve link utilization. However, the extra delay caused by the frame preemption mechanism will bring jitter to high-priority flows.
3. Proposed Technique
According to the enhanced common public radio interface specification, there are three types of flows in TSN networks:(1)IQ flow: It is the main traffic in the form of in-phase and quadrature modulation (IQ) in fronthaul transmission, and it is usually of constant rate and periodicity.(2)C&M: It refers to control and management (C&M) data. It is of an unsteady rate with no specific requirement for delay and no deterministic requirement for jitter.(3)Synchronous information flow: It is used for time synchronization of CPRI frames. It is of constant rate with low delay and no deterministic requirement for jitter.
Usually, IQ flows are classified as high priority, C&M flows, and synchronous information flows are classified as low priority or background flows. The principle of traffic scheduling is to prioritize the transmission of high-priority traffic and, at the same time, to make background flow utilize the remaining bandwidth as much as possible.
Since the number of traditional TAS scheduling queues is eight, with one flow corresponding to one queue, it is not enough for a fronthaul network with a complex traffic environment. Besides, flows with multiple functions coexisting in one queue affect normal scheduling among them. This is because background flows will affect the throughput or jitter of high-priority flows due to their nonconstant rate characteristics.
To effectively and reasonably schedule background flows, we improve the TAS scheduling mechanism to introduce a hierarchical cross-scheduling model, shown in Figure 1. In Figure 1, all incoming queues first enter the “Traffic Classification” module, then are switched to the high-priority module based on the TAS or the low-priority module based on our proposed mechanism, respectively. The high-priority module includes a high-priority queue, a GCL, and a selection transmission module. The low-priority module includes a first-stage queue, a regulator, a second-stage queue, and a queue selection module. In Figure 1, q0, q1,…, and q7 represent a total of 8 levels of high-priority flows. All other traffic is of low priority, which will be further placed in four parallel waiting queues: Q1, Q2, Q3, and Q4.
Compared to the traditional TAS mechanism, the proposed HC-TAS has at least three advantages: First, there are more high-priority queues, so the schedulability of high-priority traffic scheduling is enhanced, and the delay and jitter are reduced. Second, different queues can be selected while reducing the impact of large bursts. By reasonably adjusting the forwarding order of data frames, bandwidth utilization can be improved, and the impact on high-priority traffic can be reduced. Third, to avoid the blocking effect caused by “first come, first serve” frames, we use a hierarchical queue to schedule frames according to our designed inbound queue and outbound queue algorithms in the low-priority processing module. By doing so, it can improve bandwidth utilization.
The inbound queue and outbound queue algorithms work in the low-priority processing module and are marked in red in Figure 1. They adjust the forwarding order of background flows. Figure 2 shows the internal structure of the low-priority processing module. Taking the GCL list and background data frames in buffer S1 as input, the submodule “token control and queue selection” uses an inbound queue algorithm to select a queue that can maximize the utilization of the available time slot. The regulator is used to process burst flows. The submodule “queue status maintenance and forwarding” uses an outbound scheduling algorithm to select a data frame from one of the queues (Q1–Q4) for forwarding with polling mode.
This mechanism performs scheduling in the low-priority queue according to the GCL list, frame size, and the queue status; thus, it conforms to the nonconstant rate characteristics of low-priority traffic. The inbound queue scheduling is based on frame length matching, and its details are introduced in Algorithm 1. The outbound queue scheduling algorithm is based on queue status, and its details are introduced in Algorithm 2. Key parameter symbols in Algorithm 1 and Algorithm 2 are shown in Table 1.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||
4. Experimental Results and Analysis
For TSN in 5G networks, there are typically three scheduling modes in the mixed environment of nonconstant rate flows, and IQ flows, namely, ATS + Preemption, TAS (including guard bandwidth), and TAS + Preemption. When IQ flow is slightly complex, ATS + Preemption has a large jitter. Therefore, we only compare our proposed HC-TAS scheme with the C-FIT (one mainstream TAS scheme) and the TAS + Preemption scheme. Below, we will refer to them as scheme HC-TAS, C-FIT, and Preemption, respectively.
We simulate a fronthaul network conforming to IEEE 802.1CM. The topology is shown in Figure 3. In Figure 3, a circle without an outer contour represents the node that generates a flow, while a circle with the same filling color and a red outer contour represents the receiving node of the flow. For example, P0, P1, and P2 are sending nodes. PE is their receiving node. Table 2 summarizes all the sending nodes (source nodes) and corresponding receiving nodes (destination nodes). All rectangles designate switches, among which S1–S5 are the central switches. The topology looks like a hybrid tree, consisting of a trunk (central switches), branches (Sv, Sw, Sx, Sy, and Sz), and leaves (nodes).
Two types of flows are available in the network, namely, CBR-type IQ flow and VBR-type background flow. Each sending node generates two types of streams in the same direction. Each link is a 10 Gbps full-duplex. The IQ flow is 1 Gbps, the frame sizes are 1,500 bytes, and the transmission interval is 12,000 ns.
4.1. Experimental Results of End-to-End Delay
Figure 4 depicts the maximum end-to-end delay of all IQ flows. Our HC-TAS has the best performance. The HC-TAS usually requires more resources and might get blocked longer at one hop, but it is more advantageous in a decentralized network environment with more stream aggregation. The preemption scheme blocks a large amount of IQ data due to unpredictable preempt action, so it causes the increase of end-to-end delay of IQ flows. Since the C-FIT ensures low jitter by sacrificing latency performance, its maximum end-to-end delay of all IQ flows is longer.
4.2. Experimental Results of Jitter
Figure 5 shows the jitter of each flow. It is seen that both HC-TAS and C-FIT can reach zero jitters. Figure 5 also shows that the performance of the Preemption scheme is unstable, i.e., the jitter of each flow is different due to unstable transmission and unpredictable schedule action. Figures 4 and 5 show that our HC-TAS not only provides relatively low latency but also achieves lower jitter.
4.3. Experimental Results of Link Utilization
Now, let’s assess the relationship between maximum link utilization and jitter. In this experiment, there are more than 2,000 IQ data flow rates, randomly generated from 0.5 to 2 Gbps. Figure 6 illustrates the relation between jitter and maximum link utilization (note that the vertical axis is a logarithmic scale). As the maximum link utilization increases, the jitter of all three schemes increases. However, for the same maximum link utilization, the Preemption scheme has much more severe jitter. Both HC-TAS and C-FIT have low jitter, and the sum of jitter of HC-TAS remains more concentrated than C-FIT.
5. Conclusions and Future Work
To improve the schedulability and bandwidth utilization of background flows in TSN’s fronthaul network, we first proposed a hierarchical cross-traffic scheduling mechanism (HC-TAS) by modifying the structure of traditional TAS. Then, for the background flow, we designed an inbound scheduling algorithm based on frame length matching and an outbound scheduling algorithm based on queue status. The performance of the HC-TAS scheduling mechanism is verified through the simulation experiments. Compared with the mainstream schemes, C-FIT and Preemption, our proposed HC-TAS not only provides relatively low latency but also achieves lower jitter, and its sum of jitter remains more concentrated.
However, some future works still need to be done. The two queues in our proposed HC-TAS model are used to cache background flows. Although the C&M flow and synchronous information flow in the mobile fronthaul network belong to the same background flow, there are still differences. For example, although the C&M flow is not a high-priority flow, it includes a medium-priority part and a low-priority part. How to carry out more detailed and effective differential scheduling will be our future work.
Data Availability
The data used to support the findings of this study are available from the corresponding author upon request.
Conflicts of Interest
The authors declare that there are no conflicts of interest regarding the publication of this paper.
Acknowledgments
This research was funded by the Science and Technology Project of the State Grid Corporation of China, with grant number 5108-202222039A-1-1-ZN.