Universal Scaling Controller for Network Functions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network function (NF) scaling methods rely on CPU load, which is unreliable due to high-frequency polling routines, leading to inaccurate overload or underload detection, resulting in increased latency or packet drops.
Innovation Solution
A universal scaling controller (USC) that monitors packet distributions and queue occupancy to determine the optimal number of NF instances, reinitializing distributions when changes occur, ensuring accurate estimation of maximum safe packet rates without requiring knowledge of NF internal functioning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If CPU load monitoring is used to detect NF overload or underload, then scaling decisions can be made, but the CPU load metric becomes unreliable due to high-frequency polling routines causing 100% utilization readings
Solution Approach 1:
The patent extracts the scaling decision-making function from the NF instances themselves and relocates it to a separate scaling controller. The scaling controller monitors NF performance metrics (such as packet processing rates, queue depths, or throughput) independently of CPU load, allowing accurate overload detection without being affected by the polling-induced 100% CPU utilization. This separation enables reliable scaling decisions based on actual NF workload rather than CPU metric distortions.
2Reliability
If multiple NF instances are provisioned to handle overload, then packet drops and latency are reduced, but network resources are wasted when NFs are underloaded
Solution Approach 1:
The scaling controller implements continuous feedback monitoring of NF performance metrics to dynamically adjust the number of NF instances. When metrics indicate overload conditions (such as increased queue depth or decreased throughput), the controller provisions additional instances. When metrics show underload conditions, the controller deprovisions instances. This feedback mechanism ensures network performance stability while minimizing resource waste by matching NF instance count to actual workload demands.
Solution Approach 2:
The system transitions from static NF instance provisioning to dynamic scaling based on real-time performance metrics. The scaling controller continuously monitors NF workload and adjusts the number of instances accordingly, allowing the system to adapt to changing traffic conditions. This dynamic approach maintains adequate capacity during overload while reducing resource consumption during underload periods.
3Ease of operation
If a fixed number of NF instances are allocated, then resource management is simplified, but the system cannot adapt to changing traffic rates causing overload or underload states
Solution Approach 1:
The scaling controller enables the NF system to self-adjust its capacity by automatically provisioning or deprovisioning instances based on monitored performance metrics. Rather than requiring manual intervention or complex static allocation rules, the system autonomously responds to traffic changes by adjusting instance count, maintaining both operational simplicity and adaptability.
Data Source
AI summary
A method for self-monitored universal scaling of software network functions involves receiving, at a switch of a network, one or more batches of data units. The network further includes one or more network function (NF) instances of an NF service, and a scaling controller. The switch transmits first data units to an NF instance of the NF service during a first time period. A first distribution associated with the NF instance is updated using the first transmitted data units. Upon determining that the updated distribution has changed such that a first measure of the first distribution is outside of a first confidence interval threshold, the first distribution is reinitialized. The switch transmits second data units to the NF instance during a second time period. The reinitialized first distribution is updated using the second transmitted data units to produce a second distribution associated with the NF instance.