.png){kind=icon}
.png){kind=icon}
Return
ALAS: Attention mechanism-based and load-aware active scheduling for server resource optimization in complex scenarios
.png){kind=icon}
PRE
AI
S
J
X
R
李
C
Y
W
DOI：10.1016/j.displa.2026.103438.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Developing efficient resource scheduling strategies represents a critical challenge in cloud computing and distributed systems, particularly given the increasingly complex and dynamic workload patterns that are characteristic of modern data centers. However, the current approaches predominantly employ reactive scheduling mechanisms that adjust their resources only upon detecting insufficiency or overload conditions, resulting in inherent latency issues and inadequate responsiveness to rapidly evolving load dynamics. To address the fundamental challenges, including scheduling lag, insufficient heterogeneous data processing capabilities, and constraints with respect to modeling both short-term and long-term dependencies, this paper proposes an adaptive load prediction-driven resource optimization framework. Our approach introduces three key innovations. First, we develop a prediction-aware proactive scheduling mechanism that transforms traditional reactive resource management strategies into anticipatory allocation strategies. Second, we design a dual-attention architecture that combines full attention mechanisms with probabilistic attention to effectively capture both short-term load fluctuations and long-term evolutionary patterns, thereby significantly increasing the accuracy of prediction. Third, we propose a sample-aware learning mechanism featuring an asymmetric loss function that categorizes training samples into normal, hard-to-learn, and anomalous classes, implementing differentiated weighting strategies to improve the adaptability of the model across diverse data distributions. Experimental validations conducted on nine benchmark datasets demonstrate that the proposed ALAS strategy achieves significant advantages in terms of resource utilization optimization.
Keywords:
Edge computing
Resource optimization scheduling
Complex scenarios
Load prediction
Sample awareness
Journal
IF:
3.4
Papers:
2.1K
Citations:
3.2K
Organization
Cited Papers