|Speaker :||Leonardo Linguaglossa|
|Time:||2:00 pm - 3:00 pm|
|Location:||Paris-Rennes Room (EIT Digital)|
Network Function Virtualization (NFV) is a novel trend in IT networks which advocates replacing hardware middleboxes with equivalent pieces of code executed on general-purpose servers.
This on-going process of network softwarization can help network operators to reduce the operational and management costs of their network infrastructures.
Alongside the benefits from the operator’s perspective, new strategies to provide the network’s resources to users are arising.
Following the principle of “everything as a service”, multiple tenants can access the required resources — typically CPUs, NICs, or RAM — according to a Service-Level Agreement (SLA). This provides the opportunity for network managers to continuously monitor the network’s behavior, collect a huge amount of network data that can be used, in conjunction with machine learning algorithms, to solve several problems (from resource allocation, to intrusion detection/prevention).
Unfortunately, there is no free lunch, and current machine learning techniques require a huge amount of computational resources: in the case of softwarized networks, allocating the computational resources requires a tradeoff between the network communication’s capabilities (i.e., the “speed”) and the quality of the ML algorithm (i.e., the “intelligence”).
In this talk, I will present a network inference problem, which consists in identifying the global behavior of a high-level network function executed on top of a commodity server, by accessing low-level information represented by the server’s CPU utilization.
The focus of this presentation will be on the performance (accuracy) of the ML algorithm utilized to make the prediction, and on the impact of such an algorithm on the overall network performance.
I will finally identify the most promising approaches to be able to deploy machine learning algorithms together with real-world high-speed network applications.