The initial strategy for implementing QoS is profile mapping.
This strategy is based on the configuration to map QoS requirements specified in a declarative intent to a QoS profile for each technology that the system supports.
Deterministic Network Calculus
Network calculus is a theoretical framework for analysing performance guarantees in computer networks. Analysing performance guarantees using classic queuing theory is problematic because it relies on several assumptions about traffic flows that are not necessarily valid. In real networks, the exact properties of the traffic are often unknown.
Network calculus does not rely on assumptions about traffic flows therefore the result obtained can provide a better model of network behaviour.
Resource planning to achieve QoS end-to-end in a network slice is the objective of each transformed enterprise solution.
The best way to efficiently achieve an adequate level of quality for a network slice is to apply the quality of service on each domain where the network slice is provisioned.
The quality of service for a slice is constrained by the quality of service available in each domain.
Outside safety-critical applications guarantees about quality can be probabilistic to increase efficiency. Assuming a mixed service-provider and enterprise solution, a network slice will likely be provisioned through multiple network domains.
Probabilistic, Data-driven and Machine Learning
This is a future direction for the system. As the system is rolled out to enterprises the ability to collect QoS configuration and performance data for slices can be used to implement a mechanism to configure QoS based on Machine Learning (ML). In addition, we can consider some unsupervised learning techniques, such as Generative Adversal Networking (GAN), to synthesise data for training a QoS ML model.
If an implementation of deterministic network calculus yields positive but insufficient results a more complex version, probabilistic network calculus could be considered. This more complex solution is considered only when the general approach of using network calculus has been proven.
As with any statistical method, having enough of the correct data is the foundation so the first step is ensuring the information is collected and, where possible, labelled.
Monitoring and Scoring
In the context of the network “you can’t improve what you can’t measure” is a truism therefore monitoring and slice-specific measurement of network performance evaluated against the slice intent is a critical function for the Zeetta system.
The overall strategy for monitoring is to combine slice-specific active measurements and passive measurements to obtain a segmented view of performance across the slice together with the overall performance for the slice manifested as an aggregated network quality score and coarse-grained information about poorly performing parts of the network.
The scoring parameters (e.g. thresholds and gradient) are configurable and can be tuned for particular slices and applications.
In addition to the score for network quality, application experience obtained from application monitoring is also kept.
The platform is pluggable with respect to integration with monitoring systems, probes and network equipment to configure or collect data and includes support for containerised probe placement. As with all other supported technology domains, monitoring is vendor-agnostic.
Adaption and Elasticity
The final aspect of Zeetta’s QoS subsystem is adaption. Adaption is the process of modifying the network configuration of the slice to either better match the intent or better serve an application as that application’s network needs change. Adaption can be manual (a user changing the intent) or automated (automated modification of the intent or of the slice blueprint that attempts to deliver the slice).
The degree to which adaption can be automated ranges from simple rule-based automation to extremely complex, multi-parameter configuration modification powered by data and Machine Learning (ML).
Elasticity is a particular form of adaption by which the resource and capabilities of the slice are scaled up or down to optimise for quality, efficiency and cost.