.png){kind=icon}
Return
GraphDelta: A distributed incremental framework for efficient dynamic graph computing in edge intelligence
.png){kind=icon}
PRE
AI
DOI：10.1016/j.sysarc.2026.103834.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
As large language models migrate to edge intelligence, efficient processing of underlying dynamic graphs becomes vital for low-latency reasoning and resource-constrained execution. However, existing distributed systems often suffer from redundant computation and varying convergence speeds when handling dynamic graph data. To address these issues, we present GraphDelta, a unified framework that combines inter-batch and intra-batch optimizations. The inter-batch incremental update model reuses historical results and applies a pruning function to reduce the impact of vertex deletions, while the intra-batch incremental execution strategy selectively updates active vertices. Moreover, we design and implement GraphDelta based on GraphX, a widely used platform for distributed graph computing, and conduct experiments using representative benchmarks: PageRank, Connected Components, and SSSP on four different graph datasets. Experiment results indicate that GraphDelta outperforms GraphX with an average speedup of 39.11x when the graph update size is |ΔG|=100k, and exceeds the performance of other incremental graph processing systems with an average speedup of 4.96x when the graph update size is |ΔG|={1%,5%,10%,15%,20%}|G|.
Keywords:
GraphDelta
dynamic graph computing
edge intelligence
incremental processing
distributed systems
Journal
IF:
4.1
Papers:
3.0K
Citations:
4.2K
Organization
Cited Papers
.png){kind=icon}
ShareSaveShareSaveShareSave
ShareSaveA Real-Time Partition Generation Mechanism for Data Skew Mitigation in Spark Computing Environment
PREAI
ShareSaveShareSave
View more