In edge computing environments, device resources are often highly restricted. To cope with increasingly complex deep learning tasks, decomposing monolithic models into multiple collaborative microservice containers has become an effective approach to break through computing power bottlenecks. However, existing container orchestration solutions (e.g., K3s, KubeEdge) suffer from excessive control plane overhead, redundant network protocol stacks, and a lack of fine-grained flow control on the edge. Furthermore, the NAND Flash storage media widely used in edge devices face serious risks of write amplification and wear. Addressing these challenges, this paper proposes SlotFlow, a lightweight orchestration architecture based on In-Memory File-Slots. The core idea of SlotFlow is to utilize the host's tmpfs as a zero-copy communication bus, achieving microsecond-level synchronization through atomic file operations. Additionally, it introduces a backpressure mechanism based on queueing theory and control theory to effectively prevent memory overflows. Experimental results demonstrate that SlotFlow’s communication performance on a single node approaches that of Unix Domain Sockets, reducing latency by approximately 84% compared to traditional TCP loopback. Theoretically, it reduces physical Flash writes to zero. In overload scenarios, SlotFlow's backpressure mechanism reduces task queue backlog by approximately 73% and extends the system's crash-free runtime by 2.5 times compared to the control group, achieving highly robust dynamic task orchestration.

Jingyang Liu. SlotFlow: In-Memory File-Slot Orchestration and Adaptive Backpressure Mechanism for Constrained Edge Environments. Academic Journal of Computing & Information Science (2025), Vol. 8, Issue 11: 71-78. https://doi.org/10.25236/AJCIS.2025.081108.

[1] Pan, A., & Ray, P. P. (2025). LLMYOLOEdge: An edge-IoT aware novel framework for integration of YOLO with localized quantized large language models. IEEE Access, 13, 167250–167279. https://doi.org/10.1109/ACCESS.2025.11176056

[2] Li, H., Li, X., Fan, Q., He, Q., Wang, X., & Leung, V. C. M. (2024). Distributed DNN inference with fine-grained model partitioning in mobile edge computing networks. IEEE Transactions on Mobile Computing, 23(10), 9060–9074. https://doi.org/10.1109/TMC.2024.3357346

[3] Yakubov, D., & Hästbacka, D. (2025). Comparative analysis of lightweight Kubernetes distributions for edge computing: Performance and resource efficiency. In Service-Oriented and Cloud Computing. ESOCC 2025 (pp. 1–22). Springer. https://www.researchgate.net/publication/390570884_ Comparative_Analysis_of_Lightweight_Kubernetes_Distributions_for_Edge_Computing_Performance_and_Resource_Efficiency

[4] Alqaisi, O. I., Tosun, A. Ş., & Korkmaz, T. (2024). Performance analysis of container technologies for computer vision applications on edge devices. IEEE Access, 12, 41852–41869. https://doi.org/10.1109/ACCESS.2024.3375836

[5] Gupta, R., & Nahrstedt, K. (2025). Performance characterization of containers in edge computing (arXiv:2505.02082). arXiv. https://arxiv.org/abs/2505.02082

[6] Oh, S., Kim, J., Han, S., Kim, J., Lee, S., & Noh, S. H. (2024). MiDAS: Minimizing write amplification in log-structured systems through adaptive group number and size configuration. In Proceedings of the 22nd USENIX Conference on File and Storage Technologies (FAST 24) (pp. 1–16). USENIX Association. https://www.usenix.org/conference/fast24/presentation/oh

[7] Ekane, B., Mvondo, D., Lachaize, R., & Bromberg, Y.-D. (2025). DiSC: Backpressure mitigation in multi-tier applications with distributed shared connection. In Proceedings of the 22nd USENIX Symposium on Networked Systems Design and Implementation (NSDI 25) (pp. 55–70). USENIX Association. https://www.usenix.org/conference/nsdi25/presentation/ekane

[8] Matsubara, Y., Levorato, M., & Restuccia, F. (2023). Split computing and early exiting for deep learning applications: Survey and research challenges. ACM Computing Surveys, 55(5), Article 90. https://doi.org/10.1145/3527155

[9] Sedlak, B., Raith, P., Morichetta, A., Casamayor Pujol, V., & Dustdar, S. (2025). Multi-dimensional autoscaling of stream processing services on edge devices (arXiv:2510.06882). arXiv. https://arxiv.org/ abs/2510.06882