Split Learning (SL) enables collaborative model training by distributing computation between resource-constrained mobile edge devices and a central server. In this distributed framework, mobile edge devices compute the early layers of the model and transmit feature maps to the server. However, due to practical mobile wireless bandwidth limitations, these feature maps should be compressed before transmission. In this paper, we propose a lightweight Mask-Encoded Sparsification (MS) compressor that enhances Top-<inline-formula><tex-math notation="LaTeX">$k$</tex-math><alternatives><mml:math><mml:mi>k</mml:mi></mml:math><inline-graphic xlink:href="qu-ieq1-3696694.gif"/></alternatives></inline-formula> sparsification with a narrow-bit mask, significantly reducing compression error while improving computational and communication efficiency. Our theoretical analyses reveal that feature map compression generally induces estimator bias in the gradients. Furthermore, we establish the relationships between compression error and convergence, as well as between the choice of cutlayer and output error. The first relationship demonstrates that reducing compression error facilitates convergence, confirming the advantage of MS, whereas the second provides theoretical guidance for determining the cutlayer, indicating that earlier layers are more sensitive to compression errors. Extensive experiments on vision and language models demonstrate that our method outperforms existing compression techniques in terms of compression error and model performance, achieving up to a 21.38× speedup in communication efficiency under wireless environments.
Wen-Xuan Zhou, Zhi-Hao Qu, Baoliu Ye et al.· IEEE Transactions on Mobile...· 0 citations
Thanks to notable performance and capacity advantages, NVMe SSDs promise an effective cache tier for alleviating the load pressure of back-end storage servers. Compared with DRAM, an NVMe SSD exhibits up to hundreds of times larger capacity but delivers two orders of magnitude smaller bandwidth per gigabyte. This paper reveals that these hardware characteristics challenge fundamental design goals of caches: (1) storing tons of objects in a large-capacity SSD easily induces severe, persistent fragmentation, resulting in low cache space utilization; (2) intensive front-end requests contend for limited SSD bandwidth, causing unpredictable cache lookup latency. To tackle these challenges, this paper presents Gemini. The core of Gemini is heat-informed, group-based object management with a hardware-assisted I/O prioritization mechanism. Specifically, we introduce a tree-based prediction model for agile object grouping and fast reaction to hotspot shift. It provides SSD-friendly, bulk object eviction with a zero-write, remapping-based migration mechanism. In addition, to ensure request performance, we propose a novel I/O model that mitigates bandwidth congestion and prioritizes their processing at the hardware layer. Evaluation results show that Gemini delivers up to 3.8<inline-formula><tex-math notation="LaTeX">${\boldsymbol{\times}}$</tex-math><alternatives><mml:math><mml:mrow><mml:mo mathvariant="bold">×</mml:mo></mml:mrow></mml:math><inline-graphic xlink:href="cai-ieq1-3711088.gif"/></alternatives></inline-formula> throughput improvement, 2<inline-formula><tex-math notation="LaTeX">${\boldsymbol{\times}}$</tex-math><alternatives><mml:math><mml:mrow><mml:mo mathvariant="bold">×</mml:mo></mml:mrow></mml:math><inline-graphic xlink:href="cai-ieq2-3711088.gif"/></alternatives></inline-formula> tail latency reduction, and 2.7<inline-formula><tex-math notation="LaTeX">${\boldsymbol{\times}}$</tex-math><alternatives><mml:math><mml:mrow><mml:mo mathvariant="bold">×</mml:mo></mml:mrow></mml:math><inline-graphic xlink:href="cai-ieq3-3711088.gif"/></alternatives></inline-formula> higher cache space utilization than state-of-the-art caches for both YCSB workloads and production applications.
