Towards probe planning for in-band network telemetry

Abstract

INT is gaining traction as an advanced network monitoring approach. Despite a few recent initiatives to orchestrate the collection of in-band network statistics, state-of-theart approaches fall short when it comes to efficiently (i) collect telemetry items while subjected to real-world constraints and (ii) considering the possibility of device failures (e.g., power failure, hardware failure). In this research work, we introduce (i) P 2 INT and FP2 INT. P 2 INT coordinates how probing packets are generated and routed to ensure that all links are covered so that the required in-band network telemetry data is collected. We theoretically formalize the problem as a Mixed-Integer Linear Programming model and propose an efficient mathematical programming-based heuristic to solve it, namely fix-and-optimize, which iteratively chooses which model’s variables would be optimized, and which ones would be fixed (hence the name fix-and-optimize). Also, we theoretically formalize FP2 INT and propose a mechanism, namely Patcher – faulttolerant probing planning for INT, which reconstructs probing cycles affected by failure nodes and optimizes them while ensuring all non-affected links/statistics are still being traversed/collected. To solve this problem efficiently, we introduce a heuristic that wisely finds a high-quality solution. To the best of our knowledge, this is the first attempt to formally define and solve these problems. Results show that that P 2 INT outperforms the closest contender by a factor of up to 6x concerning the number of probing cycles generated, while Patcher reduces the number of probe cycles needed up to 5.5x compared to a state-of-the-art solution, while not increasing the INT collectors’ load.