Network Function Virtualization (NFV) focuses on decoupling network functions from proprietary hardware (i.e., middleboxes) by leveraging virtualization technology. Combining it with Software Defined Networking (SDN) enables us to chain network services much easier and faster. The main idea of using these technologies is to consolidate several Virtual Network Functions (VNFs) into a fewer number of commodity servers to reduce costs, increase VNFs fluidity and improve resource efficiency. However, the resource allocation and placement of VNFs in the network is a multifaceted decision problem that depends on many factors, including VNFs resource demand characteristics, arrival rate, configuration of underlying infrastructure, available resources and agreed Quality of Services (QoS) in Service Level Agreements (SLAs). This paper presents a bottom-up open-source NFV analysis platform (NFV-Inspector) to (1) systematically profile and classify VNFs based on resource capacities, traffic demand rate, underlying system properties, placement of VNFs in the network, etc. and (2) extract/calculate the correlation among the QoS metrics and resource utilization of VNFs. We evaluated our approach using an emulated virtual Evolved Packet Core platform (Open5GCore) to showcase how complex relation among various NFV service chains can be systematically profiled and analyzed.

Discovering insights about Virtual Network Function (VNFs) resource demand characteristics will enable cloud vendors to optimize their underlying Network Function Virtualization (NFV) system orchestration and dramatically mitigate CapEx and OpEx spendings. However, analyzing large-scale NFV systems, especially in mobile network environments, is a challenging task and requires tailor-made approaches for each particular application. In this demo, we showcase NFV-Inspector, an open source and extensible VNF analysis platform that is capable of systematically benchmark and profile NFV deployments. Based on its pluggable framework, NFV-Inspector classifies VNFs resource demand characteristics and correlate their Key Performance Indicators (KPIs) with system-level Quality of Service (QoS) measurements. 

Network Function Virtualization (NFV) becomes the primary driver for the evolution of 5G networks, and in recent years, Network Function Cloudification (NFC) proved to be an inevitable part of this evolution. Microservice architecture also becomes the de facto choice for designing a modern Cloud Native Network Function (CNF) due to its ability to decouple components of each CNF into multiple independently manageable microservices. Even though taking advantage of microservice architecture in designing CNFs solves specific problems, this additional granularity makes estimating resource requirements for a Production Environment (PE) a complex task and sometimes leads to an over-provisioned PE. Traditionally, performance engineers dimension each CNF within a Service Function Chain (SFC) in a smaller Performance Testing Environment (PTE) through a series of performance benchmarks. Then, considering the Quality of Service (QoS) constraints of a Service Provider (SP) that are guaranteed in the Service Level Agreement (SLA), they estimate the required resources to set up the PE. In this paper, we used a machine learning approach to model the impact of each microservice's resource configuration (i.e., CPU and memory) on the QoS metrics (i.e. serving throughput and latency) of each SFC in a PTE. Then, considering an SP's Service Level Objectives (SLO), we proposed an algorithm to predict each microservice's resource capacities in a PE. We evaluated the accuracy of our prediction on a prototype of a cloud native 5G Home Subscriber Server (HSS). Our model showed 95%-78% accuracy in a PE that has 2–5 times more computing resources than the PTE.

Industry is set to enter in a new revolution (Industry 4.0) backed by high inter-connectivity. Therefore, leveraging virtualization technology to deploy networks as virtualized network functions (VNFs) garnered attention. It helps the network operators and service providers to consolidate several VNFs on fewer of-The-shelf servers. This results in reducing the capital and operational expenditures while improving the resource efficiency. However, moving network functions from proprietary devices to standard servers comes with the profound cost of performance degradation. In order to overcome any performance issues to ensure service level agreement (SLA) requirements and before taking the solutions to real world, a sufficient verification and validation of VNFs is required. This is where Network Service benchmarking (NSB) plays a crucial role. NSB identifies any performance compromising bottlenecks by systematically evaluating the capacity of general purpose hardware resources, also know as network function virtualization infrastructure (NFVI), used to host single or multiple VNF instances. This paper presents a benchmarking methodology and framework to extract the correlation among the VNF quality of services (QoS) metrics and NFVI key performance indicators (KPls). For evaluation, VoerEir Touchstone platform is used to execute iPerf based benchmarking application to generate UDP based workload between VNFs. The results demonstrated that CPU utilization and L1-L3 cache memory are statistically correlated with packets dropped (0.43 and 0.47, respectively) and bandwidth utilization (0.99 and 0.92, respectively).

Cloud native computing paradigm allows microservice-based applications to take advantage of cloud infrastructure in a scalable, reusable, and interoperable way. However, in a cloud native system, the vast number of configuration parameters and highly granular resource allocation policies can significantly impact the performance and deployment cost of such applications. For understanding and analyzing these implications in an easy, quick, and cost-effective way, we present PerfSim, a discrete-event simulator for approximating and predicting the performance of cloud native service chains in user-defined scenarios. To this end, we proposed a systematic approach for modeling the performance of microservices endpoint functions by collecting and analyzing their performance and network traces. With a combination of the extracted models and user-defined scenarios, PerfSim can simulate the performance behavior of service chains over a given period and provides an approximation for system KPIs, such as requests' average response time. Using the processing power of a single laptop, we evaluated both simulation accuracy and speed of PerfSim in 104 prevalent scenarios and compared the simulation results with the identical deployment in a real Kubernetes cluster. We achieved ~81-99% simulation accuracy in approximating the average response time of incoming requests and ~16-1200 times speed-up factor for the simulation.

Network Function Virtualization (NFV) is an emerging technology to consolidate network functions onto high volume storages, servers and switches located anywhere in the network. Virtual Network Functions (VNFs) are chained together to provide a specific network service, called Service Function Chains (SFCs). Regarding to Quality of Service (QoS) requirements and network features and states, SFCs are served through performing two tasks: VNF placement and link embedding on the substrate networks. Reducing deployment cost is a desired objective for all service providers in cloud/edge environments to increase their profit form demanded services. However, increasing resource utilization in order to decrease deployment cost may lead to increase the service latency and consequently increase SLA violation and decrease user satisfaction. To this end, we formulate a multi-objective optimization model to joint VNF placement and link embedding in order to reduce deployment cost and service latency with respect to a variety of constraints. We, then solve the optimization problem using two heuristic-based algorithms that perform close to optimum for large scale cloud/edge environments. Since the optimization model involves conflicting objectives, we also investigate pareto optimal solution so that it optimizes multiple objectives as much as possible. The efficiency of proposed algorithms is evaluated using both simulation and emulation. The evaluation results show that the proposed optimization approach succeed in minimizing both cost and latency while the results are as accurate as optimal solution obtained by Gurobi (5%).

In the realm of cloud native environments, Ku-bernetes has emerged as the de facto orchestration system for containers, and the service mesh architecture, with its interconnected microservices, has become increasingly prominent. Efficient scheduling and resource allocation for these microservices play a pivotal role in achieving high performance and maintaining system reliability. In this paper, we introduce a novel approach for container scheduling within Kubernetes clusters, leveraging Graph Attention Networks (GATs) for representation learning. Our proposed method captures the intricate dependencies among containers and services by constructing a representation graph. The deep Q-learning algorithm is then employed to optimize scheduling decisions, focusing on container-to-node placements, CPU request-response allocation, and adherence to node affinity and anti-affinity rules. Our experiments demonstrate that our GATs-based method outperforms traditional scheduling strategies, leading to enhanced resource utilization, reduced service latency, and improved overall system throughput. The insights gleaned from this study pave the way for a new frontier in cloud native performance optimization and offer tangible benefits to industries adopting microservice-based architectures.