Date

2021-5

Author

ŞİMŞEK, GÖKSEL

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

420
views

171
downloads

Cite This

Network monitoring is one of the key aspects to ensure communication reliability in case of failures and malicious activities and has several design issues depending on the system characteristics. As traditional monitoring solutions usually rely on periodic updates between the network controller and ordinary nodes, scalability re- mains a challenge, especially for large-scale systems. A recent solution, the In-Band Network Telemetry (INT) framework, allows data packets to probe the nodes while traversing the network. Accordingly, INT allows special packets to carry accumulated performance information of multiple switches, reducing the overhead between the controller and other nodes. Even though INT may significantly reduce the communi- cation overhead, there are several design problems to achieve effective usage of the INT framework. These design problems are (i) minimization of the control overhead, (ii) guaranteeing the freshness of telemetry information and (iii) minimization of the redundancy. In this work, we formulate requirements as an optimization problem, Balanced Simple INT path generation Problem (BSIP), to generate balanced, simple INT paths. Due to the optimization problem’s search space complexity, we propose a heuristic, Graph Partitioned INT (GPINT), to find balanced paths to forward in-band telemetry information to satisfy these three requirements. Furthermore, we customize the INT framework to support custom and dynamic measurement ranges to achieve flexible monitoring. With this customization, the controller gains the ability to adapt measurement requests according to the network conditions. We present an extensive analysis of our approach, GPINT, and compare it with a recent study that uses Euler’s method for path generation. Our numerical results show that GPINT outperforms its opponent in terms of all three requirements. To verify our claims made in numerical analysis, we deploy path generator approaches on a simulation environment and test with various settings. The simulation results show the importance of the defined re- quirements and verify GPINT’s performance observed in numerical analysis. During the simulations, we realize that the INT framework is prone to packet losses and may cause partial information blackouts while obtaining a holistic view. Therefore, we propose a data recovery architecture as an auxiliary module to monitoring systems. We thoroughly test the recovery module in our simulations and measure its efficiency.

Subject Keywords

P4, Network monitoring, In-band network telemetry, Data recovery, Programmable data plane

URI

https://hdl.handle.net/11511/90804

Collections

Graduate School of Natural and Applied Sciences, Thesis