Efficient Scheduling for Multi-Job Federated Learning Systems with Client Sharing

DOI: 10.1109/dasc/picom/cbdcom/cy59711.2023.10361429 Publication Date: 2023-12-25T19:38:38Z
ABSTRACT
Federated Learning (FL) has emerged as a promising learning approch for data distributed across edge devices. Existing research mainly focuses on single-job FL systems. However, in practical scenarios, multiple jobs are often submitted simultaneously. Simply applying optimizations to multi-job systems results sub-optimal system performance. Specifically, we find considerably low resource utilization the client side due device heterogeneity. In this paper, exploit opportunities improve by sharing: (1) clients not selected one job could be allocated another job, and (2) that complete their tasks early preemptively assigned job. We propose an efficient scheduling algorithm systems, namely GMFL. This promptly assigns available soon it becomes available. To ensure training convergence, carefully select each while considering several constraints. conduct experiments using four popular models different datasets evaluate performance of proposed algorithm. Experimental show our significantly outperforms existing methods, with improvement up 2.03×.
SUPPLEMENTAL MATERIAL
Coming soon ....
REFERENCES (34)
CITATIONS (1)
EXTERNAL LINKS
PlumX Metrics
RECOMMENDATIONS
FAIR ASSESSMENT
Coming soon ....
JUPYTER LAB
Coming soon ....