Thiwanka Pathirana

Lakshani Manamperi, Disumi Pathirana, Thiwanka Pathirana et al.

Deploying large deep neural networks on memory-constrained mobile devices is a central challenge in edge ML. While compression, pruning, and quantization reduce per-parameter cost, transformer-based models remain too large for the 3.3-7.4 GB RAM envelope of commodity Android handsets. We present the DNN pipeline scheduling subsystem of CROWDio, which achieves practical ONNX inference across resource-constrained Android workers without model modification, by distributing memory pressure across devices via five mechanisms: JIT deferred partition loading, a single-partition-resident constraint, a 4-tier affinity scheduler, a zlib-compressed tensor transport, and a streaming 1:1 dependency model. Evaluated on DistilBERT (Sanh et al., 2019) (approximately 67 M parameters, SST-2) across five Android handsets over ten runs, our system holds peak per-device RSS to 43+-2 MB and limits battery draw to 50+-3 mAh per run, while streaming concurrency cuts batch latency 34% below barrier synchronisation.

Lakshani Manamperi, Disumi Pathirana, Thiwanka Pathirana et al.

Mobile Crowd Computing (MCdC) leverages the idle computational capacity of consumer smartphones to enable distributed task processing at scale; however, widespread real-world adoption remains constrained by the absence of developer-oriented frameworks capable of transparently managing device heterogeneity, fault tolerance, and connectivity volatility. This paper introduces CROWDio, a centralized MCdC platform comprising three tightly integrated subsystems: (i) a declarative SDK that abstracts distributed execution to a single function annotation, eliminating the need for explicit parallelism management; (ii) a tiered checkpointing mechanism that enables fault-tolerant task resumption under the memory and execution constraints inherent to mobile runtimes; and (iii) a pluggable multi-criteria scheduling framework driven by continuous live device telemetry, supporting interchangeable decision strategies without modification to the dispatch core. Empirical evaluation across six heterogeneous Android devices spanning CPU-bound, AI/NLP inference, and data-parallel workloads demonstrates that capability-aware adaptive scheduling reduces total execution time by up to 56.9% relative to naive round-robin dispatch, while the checkpointing subsystem incurs a bounded overhead of only 2-3 s per task regardless of checkpoint frequency. A system-wide Jain's Fairness Index of 0.889 confirms equitable and stable workload distribution across heterogeneous worker devices.