Flow control oriented forwarding and caching in cache-enabled networks

Abstract

The cache-enabled network architecture is very promising to improve the efficiency of content distribution and reduce the network congestion. In this paper, we propose a maximizing weighted throughput (MWT) algorithm for joint request forwarding, cache placement and flow control to dynamically optimize network performance. Specifically, a dual queue system that includes requests and data is established to retrieve the global content demands and traffic congestion information. In order to improve throughput performance as well as stabilize the queue, we formulate the flow-level throughput and design the request-level control policy by optimizing the throughput function and Lyapunov drift. The request forwarding policy adaptively allocates request forwarding rates for every link according to the differences among adjacent request queue backlogs. A novel cache priority function and a threshold-based request-dropping policy are used in the caching policy and flow control respectively to alleviate network overload. In addition, we prove the MWT algorithm achieves throughput near-optimal performance, and testify the dual queue system is stable by deducing the upper bound of the all queue backlogs. The experimental results verify the MWT algorithm stability and demonstrate the superiority of the MWT algorithm, compared to the state-of-the-art caching algorithms which are combined with back-pressure algorithm.

Introduction

With the rapid growth of mobile devices and wireless devices, we have seen traffic surges for working, entertainment and socializing. Notably, lots of repeated deliveries for popular contents would aggravate traffic burdens and cause huge resource wastes. Thereby, the wireless caching is developed to address the repeated deliveries for popular contents as one of the attractive candidate techniques for 5G communications (Luo et al., 2017, Poularakis et al., 2016, Long et al., 2016). Wireless caching technology allows base stations (BSs) and mobile devices to cache the popular contents, which reduces energy consumption and delivery delay. At the same time, the diversity of content distribution brought by caching brings some new challenges to the content distribution efficiency and network congestion issues in the cache-enabled network architecture, which is also drawing considerable interest from academia and industry.

Content caching can significantly relieve the heavy traffic burden on backhaul links and reduce service latency (Tao et al., 2016, Song et al., 2017, Shanmugam et al., 2013, Yang et al., 2018, Ali et al., 2020, Zhang et al., 2020). A smart collaborative video caching strategy is proposed in Tao et al. (2016) to improve energy efficiency in cognitive content-centric networking. The file download time is minimized in Shanmugam et al. (2013) by caching at the helpers and the cellular base station. Yang et al. (2018) analyzed and deduced the globally optimal probabilistic content placement strategy from the perspective of maximizing the hit probability under the constraint of document security. However, these caching strategies are designed based only on the known content requirements without regard to data forwarding issues, such as network congestion due to excessive traffic as well as link shared by multiple data objects.

In order to solve the above problems, significant research has also been done on jointly content caching and data forwarding. A novel jointly flow forwarding and rule caching decision is designed in Luo et al. (2020) for multiple user flows in software defined network. Liu and Shi (2019) proposed the online forwarding and real-time caching algorithms by jointly optimization request forwarding and data caching to minimize the average transmission cost, and utilized Lyapunov optimization to stable the network. Wang et al. (2018) designed a distributed scheme that combined request forwarding and content caching in order to reduce the traffic load, however no corresponding control measures can be taken when congestion occurred. These existing studies only relieve the traffic burden by caching and forwarding policies, but are invalid to solve the network congestion problems caused by malicious users who frequently request contents or misconfigured nodes that forward everything they receive. Therefore, when the system is on the verge of collapse, it is extremely important to take congestion control measures.

Congestion control can significantly improve the performance gain with caching in traffic load (Cui et al., 2016a, Lu et al., 2017, Li et al., 2020, Liu et al., 2019, Lee et al., 2015, Song et al., 2020, Huang et al., 2019). As one of the congestion control methods, flow control is effective to meet QoS requirements in practical applications or balance network load. A novel flow control mechanism is designed in Lu et al. (2017) to meet different QoS requirements and improve the utilization of network resources. In data enter networks, Lee et al. (2015) proposed a switch-based approach that identifies the target flow and provided different congestion control schemes using explicit congestion notification. In addition, consider a situation like this: if the transport path of a content object from its cache node to its destination node is severely congested, it is important to control traffic at the same time as forwarding requests. Therefore, it is necessary to jointly optimize content caching, request forwarding and flow control, which are intrinsically coupled.

In this work, we present a maximizing weighted throughput (MWT) algorithm for joint request forwarding, cache placement and flow control to improve the throughput performance of the network. To indirectly obtain the content demands and traffic congestion information, we design a dual queue system that controls the transmission rate of data by controlling the forwarding rate of requests. Moreover, we optimize the combination of throughput maximization and Lyapunov drift minimization to improve the throughput performance while stabilizing the queue. On the basis of optimization, we propose three policies including the request forwarding policy, request dropping policy and caching replacement policy. In the request forwarding policy, we schedule request for multiple data objects at a finer granularity so that they can be transmitted in the same slot, which can achieve better performance than the conventional back-pressure algorithm. The correspondence between data transmission rate and request forwarding rate is given with the help of virtual data. We adopt the threshold-based request-dropping policy to control the data flow, because of the data flow congestion caused by excessive requests. Under the dual queue system mode, a novel cache priority function based on request queue backlog and request forwarding rate is used in the caching replacement policy.

Our algorithm has the following three advantages. Firstly, requests injected into the network do not require any computation, i.e., the nodes do not need to calculate the available request receiver rates. Secondly, the algorithm has no need for explicitly exchanging the information of system time-varying parameters such as network congestion and link quality when it tracks the optimal solution. Finally, we provide delay guarantees for the network by deducing the upper bound of all queues. Our contribution is summarized as follows:

• We formulate the weighted throughput maximization problem and combine it with the Lyapunov drift minimization to propose a MWT algorithm. The proposed algorithm can dynamically provide the request forwarding rates, the request discarding rates of all data objects, and update cache contents of all nodes in each slot.

• We prove that the backlogs of request queues and data queues are deterministically upper bounded, the virtual data backlogs are stochastically upper bounded and the MWT algorithm achieves the throughput near-optimal performance.

• We provide some numerical experiments to show the stability of the MWT algorithm and demonstrate the superiority of the MWT algorithm over the state-of-the-art caching algorithms which are combined with back-pressure algorithm.

The remainders of this paper are organized as follows. Section 2 discusses the related work and Section 3 presents the network model and the queue system model. In Section 4, we propose the MWT algorithm and prove the boundedness of all queues and throughput near-optimal performance of MWT algorithm. Finally, simulation results and conclusion are given in Sections 5 Simulation, 6 Conclusion, respectively.