The excess buffering of packets in network elements, also referred to as bufferbloat, results in high latency. Considering the requirements of traffic generated by video conferencing systems like Zoom, cloud rendered gaming platforms like Google Stadia, or even video streaming services such as Netflix, Amazon Prime and YouTube, timeliness of such traffic is important. Ensuring low latency to IP flows with a high throughput calls for the application of Active Queue Management (AQM) schemes. This introduces yet another problem as the co-existence of scalable and classic congestion controls leads to the starvation of classic TCP flows. Technologies such as Low Latency Low Loss Scalable Throughput (L4S) and the corresponding dual queue coupled AQM, DualPI2, provide a robust solution to these problems. However, their deployment on hardware targets such as programmable switches is quite challenging due to the complexity of algorithms and architectural constraints of switching ASICs. In this study, we provide proof of concept implementations of two AQMs that enable the co-existence of scalable and traditional TCP traffic, namely DualPI2 and the preceding single-queue PI2 AQM, on an Intel Tofino switching ASIC. Given the fixed operation of the switch’s traffic manager, we investigate to what extent it is possible to implement a fully RFC-compliant version of the two AQMs on the Tofino ASIC. The study shows that an appropriate split between control and data plane operations is required while we also exploit fixed functionality of the traffic manager to support such solutions.

Novel applications and network scenarios challenge existing traffic management strategies. Hierarchical Quality of Service (HQoS) provides a fine control of resource sharing and delay, but traditional HQoS solutions have challenging complexity that prevents their deployment in the traffic management engine of high-speed switches. Programmable switches have emerged to make the packet processing pipelines flexible and reconfigurable, but their traffic management capabilities still rely on fixed functions that cannot handle the complexity of traditional HQoS approaches. In this paper, we show how the extended programmability can help in addressing this challenge by the application of a fundamentally different algorithm. To emulate HQoS behavior we extend our core-stateless resource sharing framework called Per Packet Value (PPV) with a HQoS-packet marker architecture called DeepQoS. DeepQoS can be used to mark resource sharing policies of different layers simultaneously and effectively at a single point, e.g., ensuring the fair share of a household’s traffic within an access aggregation network, while also controlling the shares of its subflows. In the PPV framework, bottleneck scheduling is very simple and is unaware of flows and policies, which are encoded to Packet Values. DeepQoS can use these existing simple PPV schedulers without any change. To demonstrate the deployability of the proposed method, we have created the DeepQoS marker implementation in DPDK while redesigned and implemented our packet scheduler called Virtual Dual Queue Core Stateless Active Queue Management (VDQ-CSAQM) on a P4-programmable switch. Using extensive measurements we demonstrate the unique capabilities of DeepQoS to realize rich and deep HQoS.

Network QoS, fairness and resource sharing control are not completely solved problems. Available solutions lack scalability due to maintaining flow state, require re-tuning if traffic changes, focus on a limited set of networking scenarios or require complex, centralized controllers and feedback loops. In this paper, we propose a core-stateless solution for closed network domains like access, enterprise and data center networks that handles resource sharing and provides guarantees for per-hop latency, independently. The proposed method enables controlled resource sharing by encoding the utility function of flows to Packet Value markings. This allows expressing resource sharing policies for all possible congestion situations, while operation is completely flow unaware inside the network. In addition, it also satisfies per-hop delay requirements for traffic flows independently. The separation of the delay requirements of the packets from their importance has not generally been possible by existing methods so far. The performance of the proposed method has thoroughly been analyzed by large number of simulations covering both static and dynamic scenarios and was implemented in a cloud-native virtual router implementing all the policies needed for a Broadband Network Gateway, showing good performance and better scalability than existing weighted queuing-based solutions.

Network QoS, fairness and resource sharing control are open challenges of network slicing and virtualization in 5G and future networks providing ultra-high speed Internet access. Traditional stateful solutions either employ the one-size-fits-all approach to provide services to end-users, regardless of the requirements of vertical services, or require real-time network monitoring and complex feedback loops for ensuring appropriate resource allocation at any time. In the past years, different core-stateless resource sharing solutions as scalable alternatives to traditional stateful approaches have recently emerged for closed networking domains like access, enterprise and data center networks. In this paper, we extend the core-stateless Per Packet Value (PPV) framework with the support of resource sharing policies across multiple layers of virtualization where policies defined by both physical and virtual network operators should be taken into account at the same time. To this end, we propose a re-marking mechanism that re-calculates packet markings during the transition from one virtualization layer to another, ensuring the desired resource sharing among end-users, network slices and physical networks without the need of real-time monitoring and complex feedback loops in the network core. To the best of our knowledge, this is the first core-stateless proposal that offers resource sharing for multiple layers of virtualization.