Multi-tenant Data Center Networks Research Articles

PROSE: Multi-round fair coflow scheduling without prior knowledge

Coflow scheduling plays a crucial role in enhancing network-level performance in datacenter applications. Existing works on coflow scheduling predominantly concentrate on clairvoyant schedulers, which assume prior knowledge of coflow sizes before transmission. However, this assumption is often unrealistic in practical scenarios like pipelined computation. Consequently, research on non-clairvoyant schedulers has garnered increasing attention. Most studies on non-clairvoyant schedulers focus on either improving efficiency or promoting fairness, with limited emphasis on striking a balance between the two objectives. Additionally, for non-cooperative environments like multi-tenant datacenter networks, it is essential to ensure strategy-proofness in coflow scheduling. Without adequate measures, tenants may manipulate their reported requirements to gain an unfair advantage in resource allocation. Ideally, in non-cooperative environments, a coflow scheduler should achieve efficiency, fairness, and strategy-proofness, all without prior knowledge of coflow sizes. However, few existing efforts have managed to simultaneously achieve these objectives.To fill this gap, we develop PROSE, a non-clairvoyant scheduler for non-cooperative environments. PROSE divides the scheduling time into multiple rounds and employs a two-stage scheduling mechanism based on these rounds. In the first stage, PROSE estimates the sizes of the coflows with unknown sizes by transmitting the probe flows, using a policy that combines coflow competition with batch processing. Once the information about coflow sizes is obtained, PROSE prioritizes coflows among tenants for bandwidth allocation in the second stage, aiming to accelerate coflow completion within each round. Furthermore, PROSE ensures optimal isolation guarantee by fairly limiting the total data transfer for each tenant and dynamically adjusts the scheduling order of coflows in a round-robin manner across multiple rounds. PROSE prevents tenants from attempting to consume excessive network resources through manipulative actions that deceive the scheduler. Trace-driven simulations demonstrate that PROSE outperforms the fair scheduler, achieving up to a 2.37× reduction in average coflow completion time, indicating substantial efficiency improvements.

RICH: Strategy-proof and efficient coflow scheduling in non-cooperative environments

Coflow scheduling can effectively improve the application performance and has been studied a lot in cooperative environments (e.g., private datacenter networks), where fairness is not the primary concern. In non-cooperative environments (e.g., multi-tenant datacenter networks), coflow scheduling should be strategy-proof; otherwise, some tenants could unfairly acquire more resources by cheating the scheduler. As minimizing coflow completion time (CCT) must prioritize coflows based on some specific rules (e.g., shortest-coflow-first, smallest-effective-bottleneck-first), tenants can raise the priority of their coflows by lying about the coflow information. Thus, it is a common belief that optimizing coflow performance can inevitably violate strategy-proofness.In this paper, we argue that the average CCT can be reduced without violating strategy-proofness. Our key insight is that prioritization can inherently achieve better CCT even without those specific rules such as smallest-effective-bottleneck-first. We propose RICH, a coflow scheduler in non-cooperative environments. At its heart, RICH splits the time into multiple rounds. In each round, RICH ensures that the total data transmitted by each tenant can provide optimal isolation guarantee. Among different rounds, RICH prioritizes coflow transmission among tenants in a round-robin manner. In this way, all tenants are fairly prioritized, and tenants do not necessarily gain more bandwidth by cheating. Extensive simulations show that RICH outperforms other strategy-proof mechanisms by up to 39.3% in terms of average CCT.

A Cloud-Optimized Transport Protocol for Elastic and Scalable HPC

Amazon Web Services (AWS) took a fresh look at the network to provide consistently low latency required for supercomputing applications, while keeping the benefits of public cloud: scalability, elastic on-demand capacity, cost effectiveness, and fast adoption of newer CPUs and GPUs. We built a new network transport protocol, scalable reliable datagram (SRD), designed to utilize modern commodity multitenant datacenter networks (with a large number of network paths) while overcoming their limitations (load imbalance and inconsistent latency when unrelated flows collide). Instead of preserving packets order, SRD sends the packets over as many network paths as possible, while avoiding overloaded paths. To minimize jitter and to ensure the fastest response to network congestion fluctuations, SRD is implemented in the AWS custom Nitro networking card. SRD is used by HPC/ML frameworks on EC2 hosts via AWS elastic fabric adapter kernel-bypass interface.

Residual Capacity-Aware Virtual Machine Assignment for Reducing Network Loads in Multi-tenant Data Center Networks

This paper proposes a residual capacity-aware virtual machine (VM) assignment scheme for multi-tenant data center networks. In multi-tenant data centers, tenants submit their resource requirements and the data centers provide VMs that are assigned to physical servers according to the requirements. These VMs communicate with each other to execute distributed processing. The performance of such distributed processing depends on the amount of traffic communicated by the VMs because the increase in traffic volume causes network congestion, which leads to packet losses and high transmission delay. Therefore, we need an appropriate VM assignment strategy that avoids the generation of network congestion in order to satisfy the requirements of the tenants. The proposed scheme performs VM assignment that reduces the network loads caused by traffic injected into data center networks, taking into account the traffic volume among VMs and the residual capacities of physical servers. Through simulation experiments, we demonstrate that the proposed scheme reduces the network loads efficiently.

Toward performance optimization with CPU offloading for virtualized multi-tenant data center networks

Network virtualization is an essential technique for data center operators to provide traffic isolation, differentiated service, and security enforcement for multi-tenant services. However, traditional protocols used in local area networks may not be applicable for data center networks due to the difference in network topology. Recent research suggests that layer-2-in-layer-3 tunneling protocols may be the solution to address the challenges. In this article, we find via testbed experiments that directly applying these tunneling protocols toward network virtualization only results in poor performance due to the scalability problems. Specifically, we observe that the bottlenecks actually reside inside the servers. We then propose a CPU offloading mechanism that exploits a packet steering function to balance packet processing among available CPU threads, thus greatly improving network performance. Compared to a virtualized network created based on VXLAN, our scheme improves the bandwidth for up to almost 300 percent on a 10 Gb/s link between a pair of tunnel endpoints.

Guaranteeing Heterogeneous Bandwidth Demand in Multitenant Data Center Networks

The ability to provide guaranteed network bandwidth for tenants is essential to the prosperity of cloud computing platforms, as it is a critical step for offering predictable performance to applications. Despite its importance, it is still an open problem for efficient network bandwidth sharing in a multitenant environment, especially when applications have diverse bandwidth requirements. More precisely, it is not only that different tenants have distinct demands, but also that one tenant may want to assign bandwidth differently across her virtual machines (VMs), i.e., the heterogeneous bandwidth requirements. In this paper, we tackle the problem of VM allocation with bandwidth guarantee in multitenant data center networks. We first propose an online VM allocation algorithm that improves on the accuracy of the existing work. Next, we develop a VM allocation algorithm under heterogeneous bandwidth demands. We conduct extensive simulations to demonstrate the efficiency of our method.