Deterministic Scheduling of Periodic Messages for Cloud RAN

Abstract

A recent trend in mobile networks is to centralize in distant data-centers the processing units which were attached to antennas until now. The main challenge is to guarantee that the latency of the periodic messages sent from the antennas to their processing units and back, fulfills protocol time constraints. We show that traditional statistical multiplexing does not allow such a low latency, due to collisions and buffering at nodes. Hence, we propose in this article to use a deterministic scheme for sending periodic messages without collisions in the network thus saving the latency incurred by buffering.We give several algorithms to compute such schemes for a common topology where one link is shared by all antennas. We show that there is always a solution when the routes are short or the load is small. When the parameters are unconstrained, and some buffering is allowed in the processing units, we propose an algorithm (PMLS) adapted from a classical scheduling method. The experimental results show that even under full load, most of the time PMLS finds a deterministic sending scheme with no latency. This paper is an extended version of a previous work presented at ICT conference [1]. arXiv:1801.07029v3 [cs.NI] 7 Jun 2019 be approximated even for very restricted classes of graphs.We introduce in Section IV a simple but very common topology, called the star routed network, that is studied in the rest of the article. In Section V, we study a variant of PALL called PAZL where the waiting times must all be zero. We provide polynomial time algorithms which works when the load or the routes are small, and an FPT algorithm with parameter the number of routes. We give experimental evidences that they find periodic assignments when the network is not too loaded. Finally in Section VI, we propose polynomial time heuristics and an exact FPT algorithm for the general PALL problem and experimentally show that they work extremely well even in loaded networks. In particular, the deterministic communication schemes we design largely outperform the traditional stochastic multiplexing with regard to latency. Related worksStatistical multiplexing even with a large bandwidth does not comply with the latency requirements of C-RAN. Therefore, the current solution [7], [8] is to use dedicated circuits for the fronthaul. Each end-point (RRH on one side, BBU on the other side) is connected through direct fiber or full optical switches. This architecture is very expensive and hardly scales in the case of a mobile network composed of about 10,000 base stations. The deterministic approach we propose has gained some traction recently: Deterministic Networking is under standardization in IEEE 802.1 TSN group [9], as well at IETF DetNet working group [10]. Several patents on concepts and mechanisms for DetNet have been already published, see for example [11], [12].

Deterministic Contention Management for Low Latency Cloud RAN over an Optical Ring

“…We also plan to deal with the case of several data-centers containing the BBUs instead of one. The algorithmic methods to find good emission timings in this generalization are more complicated and similar to what was proposed in [4], but while the load due to the C-RAN traffic is not too high it should work very well. The results obtained show that te N-GREEN network architecture has a high potential in term of minimization of the end-to-end latency, in a multi-QoS environment.…”

“…Finding good offsets for the C-RAN traffic is a hard problem even for simple topologies and without BE traffic, see [4]. In this section, we give a simple solution to this problem in the N-GREEN optical ring, and we adapt it to minimize the latency of the BE traffic.…”

“…Indeed, Deterministic Networking is one of the main method considered to reduce the end-to-end latency [8]. In a previous work [4], the authors have studied a similar problem for a star shaped network. In contrast with our previous work, finding emission timings so that different periodic sources do not use the same resource is easy in the context of the N-GREEN optical ring with a single data-center.…”

Scheduling periodic messages on a shared link

“…However, we have proven that, when parameterized by n the number of messages, the problem is FPT [7]. A slight generalization of PMA, with several contention points but each message only going through two of them, as in PMA, is NP-hard [7]. If the shared link is Fig.…”

“…Statistical multiplexing even with a large bandwidth does not satisfies the latency requirements of C-RAN [6], [7]. The current solution [8], [9] is to use dedicated circuits for the fronthaul.…”