In contemporary data center networks (DCNs), scheduling groups of parallel flows (coflows) has emerged as a critical task for improving application-level communication efficiency. Recently, research interest has shifted to hybrid-switched DCN architectures that integrate optical circuit switches (OCSs) and electrical packet switches (EPSs) to respectively manage both high-volume and low-volume traffic efficiently. To minimize the overall communication latency, it is essential to effectively coordinate coflows over hybrid network links. The complexity of this task, however, is substantially greater than that of scheduling on monolithic network links of either OCS or EPS. The complexity arises from allocating flows between OCS and EPS in a coordinated way, while considering both the reconfiguration delay of circuit switching in OCS and the bandwidth limitation of packet switching in EPS, and completing the transmission in the shortest time. The current solutions for scheduling coflows in hybrid-switched DCNs are primarily based on heuristics and lack formal performance guarantees. In this paper, we first establish a coflow scheduling framework for hybrid-switched DCNs, which can transform any given circuit schedule (denoted as SC) designed for pure OCS into a corresponding hybrid scheduling scheme SH. On this basis, we further present two approximation algorithms w.r.t SC under two primary reconfiguration models (i.e., all-stop model and not-all-stop model) of OCS to minimize the coflow completion time (CCT) in a hybrid-switched DCN. Theoretical analysis demonstrates that our algorithms can achieve the optimal traffic partitioning for hybrid switch environments w.r.t SC. Furthermore, we theoretically prove that the proposed algorithms can transform any SC for pure OCS with an approximation rate of λ into a corresponding coflow scheduling scheme SH tailored for hybrid switches with the approximation rate of λ+1. Extensive simulations utilizing Facebook data traces show that our algorithm performs well in minimizing the CCT compared to state-of-the-art schemes.

Scheduling of coflows, each a collection of parallel flows sharing the same objective, is an important task of data transmission that arises in the networks supporting data-intensive applications such as data center networks (DCNs). The hybrid switch design combining the optical circuit switch (OCS) and electrical packet switch (EPS) for transmitting high-volume and low-volume traffic separately has received considerable research attention. To support this design, efficient scheduling of coflows on hybrid network links is crucial for reducing the overall communication time. However, because it needs to consider both reconfiguration delay of circuit switching in the OCS and bandwidth limitation of packet switching in the EPS, coflow scheduling on hybrid network links is more challenging than on monotonic network links of either OCS or EPS. The existing coflow scheduling algorithms in hybrid switches are all heuristic and provide no performance guarantees. In this work, we first propose an approximation algorithm with a worst-case performance guarantee of$2\tau$, where$\tau\le N$is the maximum number of non-zero elements of each row and column of coflow’s demand matrix, for single coflow scheduling in an$N\times N$hybrid switch to minimize the coflow completion time (CCT). We then extend the algorithm for scheduling multiple coflows to minimize the total weighted CCT with a provable performance guarantee of$\mu\tau_{\max}$, where$\mu=4M\cdot\frac{w_{\max}}{w_{\min}}$,$\tau_{\max}=\max_{1\le m\le M}\tau_{m}\leq N$,$w_{\max}$and$w_{\min}$are respectively the maximum and minimum weights of the$M$coflows. Extensive simulations using Facebook data traces show that our algorithms outperform the state-of-the-art coflow scheduling schemes. Specifically, our algorithms transmit a single coflow up to 1.08$\times$faster than Solstice (hybrid switch) and 1.42$\times$faster than Reco-Sin (pure OCS), and multiple coflows up to 1.11$\times$faster than Solstice and 1.17$\times$faster than Reco-Mul$+$(pure OCS).

In shared data center networks, communications among users can be modeled as coflows, each comprising of a group of parallel data transmission flows. Efficient and fair scheduling of coflows is critical for improving both system performance and user satisfaction at the application level. Existing coflow scheduling methods maximizing efficiency (coflow completion time, CCT) and fairness (service isolation) simultaneously require prior knowledge of coflow (flow) size that is however not known before completion of coflow execution in reality, which limits their applicability. For information-agnostic scheduling, known results focus either solely on efficiency or fairness, but not both due to the hardness of achieving the desired compromise between them. In this paper, we first present an information-aware non-preemptive coflow scheduling algorithm, and show its provable long-term isolation guarantee under reasonable assumptions. We then adapt this algorithm to information-agnostic online coflow scheduling by combining limited multiplexing with Deep Reinforcement Learning (DRL) framework to achieve long-term isolation guarantee toward fair network sharing and lower average weighted CCT simultaneously. The simulation results show that our algorithm outperforms the state-of-the-art results of both fairness-optimal scheduling (NC-DRF) by 4.92in terms of average weighted CCT and performance-optimal scheduling (Aalo) in the metric of maximum normalized CCT. This fully demonstrates the superiority of our method in simultaneous optimization of efficiency and fairness for information-agnostic coflow scheduling.