Dragonfly+: Low Cost Topology for Scaling Datacenters

Network Function Virtualization (NFV) and service orchestration simplify the deployment and management of network and telecommunication services. The deployment of these services requires, typically, the allocation of Virtual Network Function -Forwarding Graph (VNF-FG), which implies not only the fulfillment of the service's requirements in terms of Quality of Service (QoS), but also considering the constraints of the underlying infrastructure. This topic has been well-studied in existing literature, however, its complexity and uncertainty of available information unveil challenges for researchers and engineers. In this paper, we explore the potential of reinforcement learning techniques for the placement of VNF-FGs. However, it turns out that even the most well-known learning technique is ineffective in the context of a large-scale action space. In this respect, we propose approaches to find out feasible solutions while improving significantly the exploration of the action space. The simulation results clearly show the effectiveness of the proposed learning approach for this category of problems. Moreover, thanks to the deep learning process, the performance of the proposed approach is improved over time.

show abstract

“…A similar assumption of loss rate can be found in [14]. In fact, latency and traffic traversing through the link have a non-linear relation [15].…”

Section: A Mathematical Optimisation-based Approachesmentioning

confidence: 67%

“…However, it is non-trivial to model QoS metrics such as loss rate and latency. For instance, the latency of a link depends on the traffic traversing it and the relation between traffic and latency is not linear [15]. Therefore, we cannot model the problem and solve it as an ILP.…”

Section: Thus We Havementioning

confidence: 99%

A Deep Reinforcement Learning Approach for VNF Forwarding Graph Embedding

Quang

Hadjadj‐Aoul

Outtagarts

2019

IEEE Trans. Netw. Serv. Manage.

121

View full text Add to dashboard Cite

show abstract

“…Designed for large parallel workloads, Niagara has a low-latency high-bandwidth Mellanox EDR InfiniBand interconnect in a "Drag-onfly+" topology [11,18] with 4 wings. A connectivity diagram is presented in Fig.…”

Section: Resulting Hardware Specificationsmentioning

confidence: 99%

“…Connectivity diagram of Niagara's network interconnect, of Mellanox EDR InfiniBand in a "Dragonfly+" topology[11,18] with 4 wings.…”

mentioning

confidence: 99%

Deploying a Top-100 Supercomputer for Large Parallel Workloads

Ponce

Zon

Northrup

et al. 2019

Proceedings of the Practice and Experience in Advanced Research Computing on Rise of the Machines (Learning)

146

View full text Add to dashboard Cite

Niagara is currently the fastest supercomputer accessible to academics in Canada. It was deployed at the beginning of 2018 and has been serving the research community ever since. This homogeneous 60,000-core cluster, owned by the University of Toronto and operated by SciNet, was intended to enable large parallel jobs and has a measured performance of 3.02 petaflops, debuting at #53 in the June 2018 TOP500 list. It was designed to optimize throughput of a range of scientific codes running at scale, energy efficiency, and network and storage performance and capacity. It replaced two systems that SciNet operated for over 8 years, the Tightly Coupled System (TCS) and the General Purpose Cluster (GPC) [13]. In this paper we describe the transition process from these two systems, the procurement and deployment processes, as well as the unique features that make Niagara a one-of-a-kind machine in Canada.

show abstract

“…A number of similar topologies attempting to solve some of the complexityrelated issues have been proposed, with fat tree [25], [26] being the most common of them (along with some variations -see, e.g. [27] -flattened fat trees or [28], [29] numerous variants of butterfly networks). Regardless of which class of equipment is used, the resulting multiple tree topologies are substantially redundant, as they tend to provide full mesh connectivity and an equal level of service to all nodes.…”

Section: Data Center Interconnect Networkmentioning

confidence: 99%

Energy Aware Data Centers and Networks: a Survey

Arabas

2018

JTIT

View full text Add to dashboard Cite

The past years have brought about a great variety of clusters and clouds. This, combined with their increasing size and complexity, has resulted in an obvious need for power-saving control mechanisms. Upon presenting a basis on which such solutions - namely low-level power control interfaces, CPU governors and network topologies – are constructed, the paper summarizes network and cluster resources control algorithms. Finally, the need for integrated, hierarchical control is expressed, and speciﬁc examples are provided.

show abstract

Dragonfly+: Low Cost Topology for Scaling Datacenters

Cited by 59 publications

References 2 publications

A Deep Reinforcement Learning Approach for VNF Forwarding Graph Embedding

A Deep Reinforcement Learning Approach for VNF Forwarding Graph Embedding

Deploying a Top-100 Supercomputer for Large Parallel Workloads

Energy Aware Data Centers and Networks: a Survey

Contact Info

Product

Resources

About