











摘" 要: 隨著移動互聯(lián)網(wǎng)的發(fā)展,終端業(yè)務(wù)對時延和算力的要求越來越高,采用異構(gòu)處理器構(gòu)建邊緣集群成為解決通用芯片算力不足的可行方案。然而現(xiàn)有的任務(wù)調(diào)度研究往往只考慮CPU、內(nèi)存等通用計算資源,缺少對異構(gòu)計算技術(shù)與邊緣計算相融合場景的考慮。針對邊緣側(cè)的異構(gòu)在線任務(wù)調(diào)度問題,結(jié)合時延和負(fù)載均衡兩個指標(biāo),提出一個異構(gòu)資源匹配度的概念,建立一種算力、需求和匹配度模型,并基于此設(shè)計一個在線任務(wù)調(diào)度算法。仿真實驗結(jié)果表明,對比現(xiàn)有算法,所提算法在不增加計算復(fù)雜度和時延的前提下,有效提升了集群負(fù)載均衡,減少了資源碎片,提高了邊緣側(cè)處理性能。
關(guān)鍵詞: 邊緣計算; 異構(gòu)集群; 任務(wù)調(diào)度; 資源匹配; 負(fù)載均衡; 異構(gòu)計算
中圖分類號: TN929.5?34; TP393.0" " " " " " " 文獻標(biāo)識碼: A" " " " " " " " " " " "文章編號: 1004?373X(2025)06?0031?08
Edge heterogeneous clusters online task scheduling based on resource matching
CHEN Jun1, WANG Xin2, ZENG Hao2, QIN Jian2
(1. Southwest China Institute of Electronic Technology, Chengdu 610036, China;
2. College of Communication Engineering, Chongqing University, Chongqing 400030, China)
Abstract: With the development of mobile internet, terminal services have increasingly requirements on latency and computing power. Using heterogeneous processor to build edge clusters has become a feasible solution to solve the lack of computing power of general?purpose chips. However, current task scheduling research often focuses on general?purpose computing resources such as CPUs and memory, overlooking the integration of heterogeneous computing technologies with edge computing. In allusion to the heterogeneous online task scheduling problem on the edge side, a concept of heterogeneous resource matching degree is proposed by combining the indicators of latency and load balancing. The simulation experimental results show that, in comparison with existing algorithms, the proposed algorithm can effectively improve cluster load balancing, reduce resource fragmentation, and enhance edge side processing performance without increasing computational complexity and latency.
Keywords: edge computing; heterogeneous cluster; task scheduling; resource matching; load balancing; heterogeneous computing
0" 引" 言
隨著智能終端的普及,產(chǎn)生的數(shù)據(jù)遠(yuǎn)超過去幾十年的總和,而受性能、能耗等因素限制,僅靠終端很難滿足用戶在算力和延遲方面的需求。為此,人們分別在云端和邊端提出了云計算和邊緣計算。其中邊緣計算作為云計算在邊緣端的延伸部分,可大幅降低時延,提高網(wǎng)絡(luò)和計算資源利用率,但是其計算資源相較云計算中心還是受限的。因此,如何為邊緣側(cè)設(shè)計合理的調(diào)度方法,優(yōu)化資源和任務(wù)配置以滿足用戶需求,成為亟待解決的問題。
此外,傳統(tǒng)的計算核心通常采用CPU,隨著摩爾定律已逼近極限,性能提升也逐漸到達(dá)了瓶頸。為應(yīng)對復(fù)雜計算場景,人們引入了GPU、FPGA、ASIC等處理器進行硬件加速,尤其在邊緣側(cè),往往由異構(gòu)計算節(jié)點共同構(gòu)成邊緣服務(wù)器集群。……