Overview
The paper addresses the challenge of efficiently routing cloudified 5G network traffic across multiple wide-area networks (WANs). As 5G networks increasingly deploy software-based network functions (NFs) and applications in cloud infrastructures, traffic traverses both operator WANs and cloud WANs, complicating the assurance of end-to-end Quality-of-Service (QoS).
Problem Statement
Cloudification introduces complexity in guaranteeing QoS for performance-sensitive 5G traffic due to:
-
Dynamic Placement: 5G network functions (NFs) and applications dynamically placed in cloud data centers (DCs) and edge sites based on resource availability.
-
Inter-domain Traffic Engineering Limitations: Current intra-domain traffic engineering (TE) approaches do not efficiently manage the complex requirements of inter-domain 5G traffic.
-
On-demand Routing Needs: Traditional TE mechanisms, operating periodically, fail to handle real-time 5G flow demands.
Solution: OTTER (Overlay Traffic Transport and Efficient Resource Allocation)
OTTER orchestrates 5G flow placement across multi-WAN overlays combining operator and cloud WANs. It has two main components:
-
OTTER Controller: Dynamically allocates network and compute resources, optimizing flow placement using real-time performance metrics (throughput, RTT, jitter, packet loss) and available compute resources.
-
OTTER Orchestrator: Implements scalable forwarding mechanisms to steer traffic along optimal multi-WAN paths, managing cloud resources and compute deployments.
Contributions
- Optimization Algorithm: OTTER employs a linear-programming-based algorithm to allocate compute and network resources efficiently to satisfy diverse 5G flow requirements.
- Dynamic, Fine-Grained Routing: Unlike traditional periodic TE solutions, OTTER dynamically adjusts flow placements, reacting to real-time network conditions and fine-grained QoS needs of different flows.
- Scalable Multi-WAN Overlay: The overlay leverages cloud-native functionalities (VMs, VPN gateways, user-defined routing) without needing access to proprietary or private network data, facilitating large-scale deployments.
Results and Evaluation
Evaluations across two commercial cloud WANs (Azure and Google Cloud) demonstrate OTTER’s effectiveness:
- Throughput: Achieved a 13% higher average throughput, with maximum improvements up to 136% (adding 6–10 Gbps).
- Latency: Reduced average round-trip time (RTT) by 15%, with maximum reductions up to 42 ms.
- Jitter and Packet Loss: Reduced jitter by 45% on average; reduced packet loss from an average of 0.06% to below 0.001%.
Additionally, OTTER’s optimization method allocated 26–45% more bytes compared to greedy baseline methods, closely approximating the performance of an ideal but practically infeasible infinitely-fast optimizer.
Conclusion
OTTER demonstrates the significant potential of multi-WAN coordination and overlay-based orchestration in next-generation cloudified 5G networks, notably surpassing traditional WAN traffic management methods.
-
Capturing Complex Network Behaviors, real-world constraints and objectives that are often nonlinear in nature.
-
adapting to rapidly changing network conditions, such variable link capacities, energy consumption, and intricate QoS requirements
-
Improved Generalization and Adaptability, generalize from historical data to make informed decisions in previously unseen scenarios.
-
large-scale NLP problems in real-time (additional computational complexity)
-
require more sophisticated implementation strategies