There is no system yet that provides predictable data plane and control plane operations in programmable networks. However, both predictable data plane and control plane operations are needed, e.g., in industrial networks. Particularly there, the operation of the network needs to be planned and, hence, relies on network operations that are deterministic and executed in a timely manner. To fill this gap, this paper proposes our system namedNAGA, which provides data plane deterministic guarantees along with consistent and timely network updates in programmable networks. In order to not rely on specialized hardware,NAGAuses widely-available hardware capabilities such as priority queuing and label-based forwarding. Whereas the real implementation ofNAGAin a P4-based testbed demonstrates that applications receive guaranteed performance in terms of latency and data rate, simulation studies show the ability ofNAGAto be even deployed in large scale scenarios beyond industrial networks, such as wide area and data center networks.

Airline companies are currently investigating means to improve in-flight services for passengers. Given emerging Air-to-Ground (A2G) communication technologies and the high desire of passengers for in-flight services, the servers providing in-flight services can be moved from the airplane to Data Centers (DCs) on the ground. In this scenario, network nodes (airplanes) demanding network services move over a ground core network. Therefore, the selection of DCs to connect to, as well as the underlying routing decisions are challenging. In particular, to keep a low-delay in-flight connection during the flight, airplanes connections can be reconfigured from a DC to another one, which comes at a delay cost. This paper presents a formal model for the in-flight service provisioning problem, also as an Integer Linear Program (ILP). We show that the problem is NP-hard and hence propose an efficient online heuristic, HOMA, which addresses the above challenges in polynomial time. HOMA models the problem as a dynamic matching with special properties, and then efficiently solves it by a transformation into the shortest-path routing problem. Our simulation results indicate that HOMA can achieve near-optimal performance and outperform the baseline and state-of-the-art algorithms by up to 15% while reducing the runtime from hours to seconds.

To provide the predictability required by emerging applications, operators typically rely on policing and/or shaping at the edge to ensure that tenants do not use excess bandwidth that was not accounted for. One of the promises of 6G is to deploy applications with strict predictability requirements across subnets and even over the Internet, where policing cannot be implemented in the end hosts. This paper presents an empirical study of the ability of modern programmable network devices to implement predictable traffic policing in the network. We find out that none of the five investigated hardware switches can provide accurate traffic policing, a key requirement for providing predictable service to applications. We observe that the switches let applications send more than what they should be allowed to, reaching up to 60% and 100% relative error for the rate and burst parameters. We further uncover the fact that switches cannot police arbitrarily low bursts, e.g., not less than 13 kilobit for one of our switches. We investigate how such limitations impact the performance of state-of-the-art solutions for predictable latency such as Chameleon. We observe that, for ensuring its predictable guarantees, Chameleon rejects around 50% of the tenants it could accommodate if switches were perfect, hence decreasing by the same ratio the revenue for the operator. Based on these observations, we discuss solutions toward more accurate and predictable policing in wide-area networks.

Providing predictable performance to tenants is mission critical for network hypervisors. As a hypervisor acts as an intermediary between tenants controllers and the physical infrastructure, its resources (e.g., CPU, RAM) should be provisioned and allocated carefully. Initially, we demonstrate that state-of-the-art CPU prediction approaches are not suitable for provisioning network hypervisor CPU resources, since they predict only the mean CPU utilization. However, provisioning the resources with a mean value can significantly degrade the forwarding performance of a network hypervisor. In this article, we present a novel approach which provisions network hypervisor CPU resources efficiently, while avoiding performance degradation. We take three steps to achieve our goal:(i)conducting a profound measurement campaign to determine what is the minimum amount of CPU resources that needs to be allocated to a network hypervisor in order to have no performance degradation;(ii)revealing the key properties of virtual networks that affect the CPU utilization;(iii)designing a precise CPU prediction model. Using randomly generated virtual networks and arbitrary physical topologies, we show that our prediction model exhibits an average relative error of around 4%. Further, our evaluations indicate that provisioning the CPU resources of a network hypervisor based on the proposed prediction model does not degrade the hypervisor forwarding performance. Utilizing our approach, network operators can minimize their resources consumption while still providing predictable and undegraded forwarding performance to tenants.

Service function chains (SFCs) are an ordered set of virtual network functions (VNFs) which can realize a specific network service. Enabled by virtualization technologies, these VNFs are hosted on physical machines (PMs), and interconnected by network switches. In today networks, these resources are usually under-utilized and/or over-provisioned, resulting in power-inefficient deployments. To improve power-efficiency, SFCs should be deployed utilizing the minimum number of PMs and network equipment, which are not concomitant. Considering the existing PM and switch power consumption models and their resource constraints, we formulate the power-aware and delay-constrained joint VNF placement and routing (PD-VPR) problem as an Integer Linear Program (ILP). Due to the NP-completeness of the problem, we proposeHolu, a fast heuristic framework that efficiently solves the PD-VPR problem in an online manner. Specifically,Holudecomposes the PD-VPR into two sub-problems and solve them sequentially:i)aVNF placementproblem that consists of mapping the VNFs to PMs using a centrality-based ranking method, andii)aroutingproblem that efficiently splits the delay budget between consecutive VNFs of the SFC, and finds a Delay-Constrained Least-Cost (DCLC) shortest-path through the selected PMs (hosting VNFs) using the Lagrange Relaxation based Aggregated Cost (LARAC) algorithm. Our simulation results indicate thatHoluoutperforms the state-of-the-art algorithms in terms of total power consumption and acceptance rate by 24.7% and 31%, respectively.

This paper presents Chameleon, a cloud network providing both predictable latency and high utilization, typically two conflicting goals, especially in multi-tenant datacenters. Chameleon exploits routing flexibilities available in modern communication networks to dynamically adapt toward the demand, and uses network calculus principles along individual paths. More specifically, Chameleon employs source routing on the "queue-level topology", a network abstraction that accounts for the current states of the network queues and, hence, the different delays of different paths. Chameleon is based on a simple greedy algorithm and can be deployed at the edge; it does not require any modifications of network devices. We implement and evaluate Chameleon in simulations and a real testbed. Compared to state-of-the-art, we find that Chameleon can admit and embed significantly, i.e., up to 15 times more flows, improving network utilization while meeting strict latency guarantees.

Today, on-board passengers desire to have in-flight services such as Voice-over-IP (VoIP) and video streaming. These services are usually hosted by geographically distributed Data Centers (DCs) that are built/rented by the airline companies. Flights can be connected to these DCs using two types of Air-to-Ground (A2G) communication alternatives: i) satellite (SC), and ii) Direct-Air-to-Ground connections (DA2G). These two options are different in terms of propagation delay, capacity, and availability. Focusing on reducing the delay of the inflight services, each airplane should be assigned to a nearby DC. However, due to the mobility of flights, a permanent DC assignment might not lead to an acceptable service delay for the flight duration. Therefore, the flight needs to be reassigned to another DC (reconfiguration) along its route, which comes with a cost. The real challenge in this work is to find the best assignments of each airplane to DC(s) and determine the required reconfigurations such that the sum of routing and reconfiguration delay is minimized. We model this problem as a Multi-Period Generalized Assignment Problem (MPGAP) and formulate it as an Integer Linear Programming (ILP) optimization model. To overcome the scalability issues of the ILP, we propose Figo, a flight control framework that solves the MPGAP problem using deep Q-learning. Considering a realistic European-based Space-Air-Ground-Integrated Network (SAGIN) and a real set of flights, we compare the performance of Figo against the optimal. The results indicate that Figo can achieve 7% optimality gap in the worst case, while reducing the runtime from hours to seconds.

To meet their increasingly stringent dependability requirements, communication networks need to be predictable, both in terms of correctness and performance. In principle, Software-Defined Networks (SDN) enable such more predictable networks, however, these networks still depend the underlying switches. This paper presents an empirical study of the predictability of SDN switches. Our extensive benchmarking of seven hardware OpenFlow switches from four different manufacturers raises several concerns regarding the dependability of these switches. We uncover several incorrect and unpredictable behaviors and performance issues. In particular, we identify unpredictable behaviors related to the management of flows and buffers, and observe that existing quality-of-service mechanisms, such as priority queuing, introduce unexpected overheads. The latter, in turn, can lead to violations of latency guarantees. Based on our insights, we discuss first solutions toward more predictable architectures.

In order to scale web or other services, the load on single instances of the respective service has to be balanced. Many services are stateful such that packets belonging to the same connection must be delivered to the same instance. This requires stateful load balancers which are mostly implemented in software. On the one hand, modern packet processing frameworks supporting software load balancers, such as the Data Plane Development Kit (DPDK), deliver high performance compared to older approaches. On the other hand, common Network Interface Cards (NICs) provide additional matching capabilities that can be utilized for increasing the performance even further and in turn reduce the necessary server resources. In fact, offloading the packet matching to hardware can free up CPU cycles of the servers. Therefore, in this work, we propose the Hybrid NIC-offloading Load Balancer (HNLB), a high performance hybrid hardware-software load balancer, utilizing the NIC-offloading hardware matching capabilities. The results of our performance evaluations show that the throughput using NIC offloading can be increased by up to 50%, compared to a high performance software-only implementation. Furthermore, we investigated the limitations of our proposed approach, e.g., the limited number of possible concurrent connections.

People desire to be connected, no matter where they are. Recently, providing Internet access to on-board passengers has received a lot of attention from both industry and academia. However, in order to guarantee an acceptable Quality of Service (QoS) for the passenger services with low incurred cost, the path to route the services, as well as the datacenter (DC) to deploy the services should be carefully determined. This problem is challenging, due to different types of Air-to-Ground (A2G) connections, i.e., satellites and Direct Air-To-Ground (DA2G) links. These A2G connection types differ in terms of cost, bandwidth, and latency. Furthermore, due to the flights' movements, it is important to consider adapting the service location accordingly. In this work, we formulate two Mixed Integer Linear Programs (MILPs) for the problem of Joint Service Placement and Routing (JSPR): i) Static (S-JSPR), and ii) Mobility-Aware (MA-JSPR) in Space-Air-Ground Integrated Networks (SAGIN), with the objective of minimizing the total cost. We compare S-JSPR and MA-JSPR using comprehensive evaluations in a realistic European-based SAGIN. The obtained results show that the MA-JSPR model, by considering the future flight positions and using a service migration control, reduces the long-term total cost notably. Also, we show S-JSPR benefits from a low time-complexity and it achieves lower end-to-end delays comparing to MA-JSPR model.

With a centralized control over the forwarding devices and the embedded flows, Software Defined Networking promises to increase the flexibility of communication networks. Meanwhile, a dynamic control plane would adapt itself in a timely manner to sustain flow setup performance in the face of traffic variations. Such adaptation depends on a careful decision of the controller placement, which is challenging because we need to consider two contradictory objectives, namely the cost of operating the control plane and the cost of its adaptation. In this work, we model the problem of operating the control plane as a multi-period offline optimization problem to minimize the total cost induced by the flow setup performance and the control plane adaptation. We leverage the lookahead control scheme and decompose the intractable offline problem into smaller instances, which are solved in an online fashion efficiently with an algorithm based on simulated annealing. We perform extensive simulations on real world topologies and show that our proposed algorithm can reduce the total cost by up to 20% compared with the reference algorithms. Further, we analyze the need of frequent control plane adaptation, and compare different control plane design choices according to a novel flexibility measure.

The Network Function Virtualization (NFV) is very promising for efficient provisioning of network services and is attracting a lot of attention. NFV can be implemented in commercial off-the-shelf servers or Physical Machines (PMs), and many network services can be offered as a sequence of Virtual Network Functions (VNFs), known as VNF chains. Furthermore, many existing network devices (e.g., switches) and collocated PMs are underutilized or over-provisioned, resulting in low power-efficiency. In order to achieve more energy efficient systems, this work aims at designing the placement of VNFs such that the total power consumption in network nodes and PMs is minimized, while meeting the delay and capacity requirements of the foreseen demands. Based on existing switch and PM power models, we formulate an Integer Linear Programming (ILP) model to find the optimal solution. We also propose a heuristic based on the concept of Blocking Islands (BI), and a baseline heuristic based on the Betweenness Centrality (BC) property of the graph. Both heuristics and the ILP solutions have been compared in terms of total power consumption, delay, demands acceptance rate, and computation time. Our simulation results suggest that BI-based heuristic is superior compared with the BC-based heuristic, and very close to the optimal solution obtained from the ILP in terms of total power consumption and demands acceptance rate. Compared to the ILP, the proposed BI-based heuristic is significantly faster and results in 22% lower end-to-end delay, with a penalty of consuming 6% more power in average.

In the past few years, datacenter (DC) energy consumption has become an important issue in technology world. Server consolidation using virtualization and virtual machine (VM) live migration allows cloud DCs to improve resource utilization and hence energy efficiency. In order to save energy, consolidation techniques try to turn off the idle servers, while because of workload fluctuations, these offline servers should be turned on to support the increased resource demands. These repeated on-off cycles could affect the hardware reliability and wear-and-tear of servers and as a result, increase the maintenance and replacement costs. In this paper we propose a holistic mathematical model for reliability-aware server consolidation with the objective of minimizing total DC costs including energyand reliability costs. In fact, we try to minimize the number of active PMs and racks, in a reliability-aware manner. We formulate the problem as a Mixed Integer Linear Programming (MILP) model which is in form of NP-complete. Finally, we evaluate the performance of our approach in different scenarios using extensive numerical MATLAB simulations.