Πανεπιστήμιο Μακεδονίας
Leveraging programmable data planes for lightweight in-band network telemetry and scalable source-routing fabrics
Abstract
dc:descriptionThe advent of 5G technology promises to deliver ultra-fast, low-latency connectivity, paving the way for new applications such as autonomous vehicles, remote surgery, and immersive augmented reality experiences. Concurrently, the rise of Software-Defined Networking (SDN) and Network Function Virtualization (NFV) has introduced a new paradigm of network management, allowing for greater flexibility, centralized control, and the ability to rapidly deploy and scale network services. In addition, advancements in network programmability, materialized through domain-specific languages such as P4, enable fine-grained control over network behavior, facilitating the deployment of innovative protocols and services without the need for replacement of infrastructure.In this respect, this thesis investigates the emerging challenges in modern networks and the increasing demand for networking resources, which impose severe requirements on the data plane. Virtualized Network Function (VNF) chains significantly strain cloud infrastructures due to their high demand for forwarding state. Routing traffic through multiple VNFs necessitates numerous flow entries across the network. In today’s multi-tenant environments, traditional switches with layer-2/3 forwarding face significant limitations. To address these challenges, we present a practical implementation of source routing using OpenFlow-enabled switches. Our comparative study between source routing and L2 forwarding, conducted using our implementation in Mininet, uncovers substantial savings in forwarding state and hassle-free network updates when employing the proposed source routing scheme.We delve deeper into network programming technologies, utilizing the P4 language to explore its benefits on In-band Network Telemetry (INT). INT enables fine-grained network monitoring and telemetry analysis in the data plane, avoiding the overhead of a centralized controller. However, existing INT frameworks introduce substantial transmission overhead, which grows linearly with the number of hops and telemetry variables. To address this, we propose a lightweight INT framework, DLINT, which exercises per-flow aggregation of telemetry metadata. More precisely, DLINT spreads telemetry information across several packets of a flow, maintaining manageable per-packet overhead. We coordinate switches by maintaining minimal telemetry state and leveraging Bloom Filters to compress lookups. Our proposed INT framework encompasses all required functionality for continuous tracing of telemetry values over the entire duration of flows. Evaluating DLINT in a path tracing application demonstrates significant gains in header space conservation and continuous delivery of path traces, even during path updates.Furthermore, we develop a probabilistic INT method, PLINT. PLINT obviates the need for coordination among switches by empowering INT nodes to insert their own telemetry metadata based on a certain probability. PLINT employs reservoir sampling to ensure an equal distribution of telemetry information among packets. Similar to DLINT, PLINT uses an adjustable INT header space, but suffers from telemetry data redundancy due to its probabilistic nature. Our evaluation results corroborates that PLINT is inferior to DLINT in terms of conveying path trace information, but detects path updates more promptly, due to its more efficient INT header space utilization.Although these techniques enhance INT performance, they exhibit implications stemming from probabilistic elements such as Bloom Filter collisions. To alleviate these implications, we introduce Controller-Assisted Lightweight INT (CLINT), an INT method that utilizes a centralized telemetry controller with minimal intervention in order to maintain a more deterministic INT behavior. More specifically, the controller’s utilization is restricted in the initial setup of each monitored flow, after which the INT process proceeds exclusively in the data plane. Our evaluation results indicate that CLINT outperforms the previous INT techniques in terms of monitoring efficiency and network update detection, while maintaining an insignificant control communication overhead.
Degree
thesis:*- Grantor dc:publisher
- Πανεπιστήμιο Μακεδονίας
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Papadopoulos, Konstantinos
- Παπαδόπουλος, Κωνσταντίνος
Subjects
dc:subject × 11- Προγραμματισμός δικτύων
- Δικτυακή τηλεμετρία
- Programmable data plane
- P4
- In-band network telemetry
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Δίκτυα ηλεκτρονικών υπολογιστών και Επικοινωνίες
- Computer and Information Sciences
- Natural Sciences
- Computer Networks and Communications
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/57117