Robust Resource Sharing in Network Slicing via Hypothesis Testing

TL;DR

The paper tackles the challenge of satisfying SLAs for multiple network slices on shared cellular infrastructure without sacrificing isolation. It introduces a two-phase approach combining trial-based stochastic modeling (via Markov chains) with Neyman-Pearson hypothesis testing to deprioritize anomalous NSs during contention, enabling efficient resource sharing. Evaluations on real LTE traces show improved SLA robustness and reduced bandwidth needs compared with exclusive provisioning or naive sharing, demonstrating a practical tradeoff between efficiency and isolation. The work offers a scalable framework for robust network slicing that can adapt to traffic anomalies and supports future enhancements in Bayesian or worst-case detection strategies.

Abstract

In network slicing, the network operator needs to satisfy the service level agreements of multiple slices at the same time and on the same physical infrastructure. To do so with reduced provisioned resources, the operator may consider resource sharing mechanisms. However, each slice then becomes susceptible to traffic surges in other slices which degrades performance isolation. To maintain both high efficiency and high isolation, we propose the introduction of hypothesis testing in resource sharing. Our approach comprises two phases. In the trial phase, the operator obtains a stochastic model for each slice that describes its normal behavior, provisions resources and then signs the service level agreements. In the regular phase, whenever there is resource contention, hypothesis testing is conducted to check which slices follow their normal behavior. Slices that fail the test are excluded from resource sharing to protect the well-behaved ones. We test our approach on a mobile traffic dataset. Results show that our approach fortifies the service level agreements against unexpected traffic patterns and achieves high efficiency via resource sharing. Overall, our approach provides an appealing tradeoff between efficiency and isolation.

Figures (14)

• Figure 1: We consider two network functions deployed at the BS. First, the BDE observes the state $X_i(t)$ of a NS and estimates the number of PRBs $W_i(t)$ needed to meet its desired QoS. Second, the NSM receives all bandwidth demands and decides which ones to satisfy given the limited bandwidth at the BS by producing binary decisions $u_i(t)$. The length of each slot $t$ depends on the timescales supported by these two network functions.
• Figure 2: When a new NS is deployed, the trial phase is initiated to obtain a stochastic model for the normal behavior of each NS and to estimate the required provisioned bandwidth. Next, the regular phase is initiated using the previously estimated provisions and models. Whenever there is resource contention, the NSM checks if each NS behaves as expected via hypothesis testing. Anomalous NSs are deprioritized in bandwidth allocation.
• Figure 3: A small sample of a parquet file found in the dataset. Here, two lines exist with the same timestamp and $\text{direction}=1$. This indicates that two UEs were receiving traffic from the BS during that subframe.
• Figure 4: Anomalies are created by changing the MC of a NS. Here, we remove the lowest $\beta=50\%$ states of the MC. Then, each state's largest transition probability is increased so that each row of the new matrix sums to 1.
• Figure 5: The ECDF of the connected users in the regular phase.