Optimized Sampling Strategies to Model the Performance of Virtualized Network Functions

Abstract

Modern network services make increasing use of virtualized compute and network resources. This is enabled by the growing availability of softwarized network functions, which take on major roles in the total traffic flow (such as caching, routing or as firewall). To ensure reliable operation of its services, the service provider needs a good understanding of the performance of the deployed softwarized network functions. Ideally, the service performance should be predictable, given a certain input workload and a set of allocated (virtualized) resources (such as vCPUs and bandwidth). This helps to estimate more accurately how much resources are needed to operate the service within its performance specifications. To predict its performance, the network function should be profiled in the whole range of possible input workloads and resource configurations. However, this input can span a large space of multiple parameters and many combinations to test, resulting in an expensive and overextended measurement period. To mitigate this, we present a profiling framework and a sampling heuristic to help select both workload and resource configurations to test. Additionally, we compare several machine-learning based methods for the best prediction accuracy, in combination with the sampling heuristic. This work has been performed in the framework of the NG-PaaS and 5GTANGO project, funded by the European Commission under the Horizon 2020 and 5G-PPP Phase2 programmes, resp. under Grant Agreement No. 761 557 and 761 493 (http://ngpaas.eu) (https://www.5gtango.eu). This work is partly funded by UGent BOF/GOA project "Autonomic Networked Multimedia Systems".

VNF Performance modelling: From stand-alone to chained topologies

“…We use the method griddata implemented in the Python SciPy library [33]. In [9] and [25], this method provided the best accuracy for profiling stand-alone VNFs. We validate it here again for our VNF chain modelling approach.…”

“…The training and prediction process is however quite calculation intensive, therefore this method becomes less convenient at large sample sizes and high dimensional models. Previous research [25], [35]) shows that Gaussian Processes are well capable of modelling non-polynomial trends of VNF performances. Therefore we also validate it here for our VNF chain use case.…”

“…In practice, the VNF is a single Virtual Machine (VM) or container in our tests. The profiling framework we use for automated workload generation and KPI monitoring is described in [25]. While the workload is flowing through the chain, specialized probes then export metrics related to resource usage, workload and performance.…”

“…Our previous work in [9] and [25] focused on profiling the performance of single VNFs. Now we consider the performance of VNF chains, but we can still apply the same performance model if we consider the total VNF chain as a closed system.…”

Adaptive & Learning-aware Orchestration of Content Delivery Services

“…This creates a good estimation of the average performance of this service topology. To speed up the measurements, we use the automated profiling environment described in [11]. This way we can run the tests in parallel, over multiple equal compute nodes.…”