ADAPTIVE PROTOCOL SELECTION LAYER FOR CONTEXT-AWARE MULTI-PROTOCOL DATA DELIVERY IN CLOUD–EDGE SYSTEMS: A CONTEXTUAL-BANDIT FOUNDATION WITH BAYESIAN WARM-START

Roman Nazarevych, Ivan Bolesta

Abstract


Background. Protocol performance in cloud–edge systems is regime-dependent: the transport that minimizes latency, loss, or CPU cost changes with payload, rate, link type, and signal quality, and tail latency often inverts the median ranking. Static choices and hand-tuned heuristic switchers cannot track these shifts, and many nearby adaptive-selection prototypes remain deterministic scorers, not online learners. This paper establishes the theoretical foundation for an adaptive protocol selection layer that treats protocol choice at decision ticks as online decision-making under partial feedback, demonstrated on isolated proof-of-concept examples.

Materials and Methods. Protocol selection is formulated as a contextual multi-armed bandit over the five supported implementation transports and solved with LinUCB: a per-arm ridge-regression estimator with an upper-confidence-bound exploration term, computed by Gaussian elimination. The four-dimensional context couples a log-linear payload feature, normalized Wi-Fi signal and link-rate features, and a bias term. Bounded penalty and peer-relative rewards are derived from blended median/tail latency, goodput, and loss. Non-stationarity is handled by per-tick geometric decay, and a Bayesian warm-start encodes offline ridge priors in the initial state; normal per-tick decay may scale that prior before the first logged score.

Results and Discussion. The five-protocol formulation is demonstrated on isolated Raspberry Pi 5B and Pi Zero 2 W examples; the adaptive algorithm uses five protocols (gRPC, gRPC-bidi, MQTT, CoAP, HTTP). Baseline runs confirm the best transport changes with payload, host, and objective; decision logs show LinUCB concentrating on a strong context-conditioned arm in stable regimes, while a deliberately retained cold-start miss exposes early-evidence path dependence; a Bayesian warm-start improves cold-start latency and throughput.

Conclusion. The paper gives a reproducible contextual-bandit formulation with Bayesian warm-start for multi-protocol cloud–edge data delivery. It defines context, reward, and exploration sufficient for further research.

Keywords: contextual bandit, LinUCB, adaptive protocol selection, cloud–edge systems, Bayesian prior, exploration-exploitation.


Full Text:

PDF

References


[1] Nazarevych, R., & Bolesta, I. (2025). Software of Internet-accessible semiconductor laboratory. Information Systems and Networks, 17, 146–159 (In Ukrainian). https://doi.org/10.23939/sisn2025.17.146

[2] gRPC Authors. (2024). gRPC: A high-performance, open-source universal RPC framework – Documentation. Cloud Native Computing Foundation. Retrieved May 17, 2026. https://grpc.io/docs/

[3] OASIS. (2019). MQTT version 5.0 (OASIS Standard). Organization for the Advancement of Structured Information Standards.
https://docs.oasis-open.org/mqtt/mqtt/v5.0/mqtt-v5.0.html

[4] Shelby, Z., Hartke, K., & Bormann, C. (2014). The Constrained Application Protocol (CoAP) (RFC 7252). Internet Engineering Task Force. https://doi.org/10.17487/RFC7252

[5] Iyengar, J., & Thomson, M. (2021). QUIC: A UDP-based multiplexed and secure transport (RFC 9000). Internet Engineering Task Force. https://doi.org/10.17487/RFC9000

[6] Bishop, M. (2022). HTTP/3 (RFC 9114). Internet Engineering Task Force. https://doi.org/10.17487/RFC9114

[7] Naik, N. (2017). Choice of effective messaging protocols for IoT systems: MQTT, CoAP, AMQP and HTTP. In 2017 IEEE International Systems Engineering Symposium (ISSE) (pp. 1–7). IEEE.
https://doi.org/10.1109/SysEng.2017.8088251

[8] Auer, P., Cesa-Bianchi, N., & Fischer, P. (2002). Finite-time analysis of the multiarmed bandit problem. Machine Learning, 47(2–3), 235–256. https://doi.org/10.1023/A:1013689704352

[9] Lai, T. L., & Robbins, H. (1985). Asymptotically efficient adaptive allocation rules. Advances in Applied Mathematics, 6(1), 4–22.
https://doi.org/10.1016/0196-8858(85)90002-8

[10] Lattimore, T., & Szepesvári, C. (2020). Bandit algorithms. Cambridge University Press. https://doi.org/10.1017/9781108571401

[11] Bouneffouf, D., Rish, I., & Aggarwal, C. (2020). Survey on applications of multi-armed and contextual bandits. In 2020 IEEE Congress on Evolutionary Computation (CEC) (pp. 1–8). IEEE.
https://doi.org/10.1109/CEC48606.2020.9185782

[12] Sutton, R. S., & Barto, A. G. (2018). Reinforcement learning: An introduction (2nd ed.). MIT Press.
http://incompleteideas.net/book/the-book-2nd.html

[13] Slivkins, A. (2019). Introduction to multi-armed bandits. Foundations and Trends in Machine Learning, 12(1–2), 1–286.
https://doi.org/10.1561/2200000068

[14] Li, L., Chu, W., Langford, J., & Schapire, R. E. (2010). A contextual-bandit approach to personalized news article recommendation. In Proceedings of the 19th International Conference on World Wide Web (WWW '10) (pp. 661–670). ACM. https://doi.org/10.1145/1772690.1772758

[15] Chu, W., Li, L., Reyzin, L., & Schapire, R. E. (2011). Contextual bandits with linear payoff functions. In Proceedings of the 14th International Conference on Artificial Intelligence and Statistics (AISTATS) (Vol. 15, pp. 208–214). PMLR. https://proceedings.mlr.press/v15/chu11a.html

[16] Hoerl, A. E., & Kennard, R. W. (1970). Ridge regression: Biased estimation for nonorthogonal problems. Technometrics, 12(1), 55–67. https://doi.org/10.1080/00401706.1970.10488634

[17] Golub, G. H., & Van Loan, C. F. (2013). Matrix computations (4th ed.). Johns Hopkins University Press. https://doi.org/10.56021/9781421407944

[18] Abbasi-Yadkori, Y., Pál, D., & Szepesvári, C. (2011). Improved algorithms for linear stochastic bandits. In Advances in Neural Information Processing Systems (NeurIPS) (Vol. 24, pp. 2312–2320). Curran Associates. https://proceedings.neurips.cc/paper/2011/hash/e1d5be1c7f2f456670de3d53c7b54f4a-Abstract.html

[19] Garivier, A., & Moulines, E. (2011). On upper-confidence bound policies for switching bandit problems. In Algorithmic Learning Theory (ALT 2011), LNCS (Vol. 6925, pp. 174–188). Springer.
https://doi.org/10.1007/978-3-642-24412-4_16

[20] Shivaswamy, P., & Joachims, T. (2012). Multi-armed bandit problems with history. In Proceedings of the 15th International Conference on Artificial Intelligence and Statistics (AISTATS) (Vol. 22, pp. 1046–1054). PMLR. https://proceedings.mlr.press/v22/shivaswamy12.html

[21] Kazerouni, A., Ghavamzadeh, M., Abbasi-Yadkori, Y., & Van Roy, B. (2017). Conservative contextual linear bandits. In Advances in Neural Information Processing Systems (NeurIPS) (Vol. 30, pp. 3910–3919). Curran Associates. https://proceedings.neurips.cc/paper/2017/hash/bdc4626aa1d1df8e14d80d345b2a442d-Abstract.html




DOI: http://dx.doi.org/10.30970/eli.34.9

Refbacks

  • There are currently no refbacks.