科技通报
科技通報
과기통보
BULLETIN OF SCIENCE AND TECHNOLOGY
2014年
10期
178-180
,共3页
复杂网络%隐性群体%并行特征
複雜網絡%隱性群體%併行特徵
복잡망락%은성군체%병행특정
complex networks%hidden groups%parallel mode matching
现实的通信网络由多重网络组成,构成具有多维任务分配的复杂网络结构,在任务处理中会产生并行振荡,对复杂网络中的振荡抑制是提高复杂网络并行处理的重要因素。传统的并行振荡方法采用奇异值分解降维的特征匹配算法,在面对大规模复杂任务求解时产生大量的内存需求和时间损耗。提出一种基于隐性群体双模分解的并行振荡抑制算法,首先进行复杂网络多维业务并行处理模型设计,得到了复杂网络多维业务并行处理模型的指标参量体系,采用隐性群体并行特征匹配方法实现双模特征匹配并行处理。仿真实验表明,采用该算法进行复杂网络隐性群体的并行特征匹配,实现并行处理和串行处理,双模分解的时间成本及空间成本大幅降低,加速比提高2倍,有效抑制网络振荡。算法在进行复杂网络多任务并行处理中发包数量,时延和能量效率等方面具有优越性能。
現實的通信網絡由多重網絡組成,構成具有多維任務分配的複雜網絡結構,在任務處理中會產生併行振盪,對複雜網絡中的振盪抑製是提高複雜網絡併行處理的重要因素。傳統的併行振盪方法採用奇異值分解降維的特徵匹配算法,在麵對大規模複雜任務求解時產生大量的內存需求和時間損耗。提齣一種基于隱性群體雙模分解的併行振盪抑製算法,首先進行複雜網絡多維業務併行處理模型設計,得到瞭複雜網絡多維業務併行處理模型的指標參量體繫,採用隱性群體併行特徵匹配方法實現雙模特徵匹配併行處理。倣真實驗錶明,採用該算法進行複雜網絡隱性群體的併行特徵匹配,實現併行處理和串行處理,雙模分解的時間成本及空間成本大幅降低,加速比提高2倍,有效抑製網絡振盪。算法在進行複雜網絡多任務併行處理中髮包數量,時延和能量效率等方麵具有優越性能。
현실적통신망락유다중망락조성,구성구유다유임무분배적복잡망락결구,재임무처리중회산생병행진탕,대복잡망락중적진탕억제시제고복잡망락병행처리적중요인소。전통적병행진탕방법채용기이치분해강유적특정필배산법,재면대대규모복잡임무구해시산생대량적내존수구화시간손모。제출일충기우은성군체쌍모분해적병행진탕억제산법,수선진행복잡망락다유업무병행처리모형설계,득도료복잡망락다유업무병행처리모형적지표삼량체계,채용은성군체병행특정필배방법실현쌍모특정필배병행처리。방진실험표명,채용해산법진행복잡망락은성군체적병행특정필배,실현병행처리화천행처리,쌍모분해적시간성본급공간성본대폭강저,가속비제고2배,유효억제망락진탕。산법재진행복잡망락다임무병행처리중발포수량,시연화능량효솔등방면구유우월성능。
The reality of the communication network composed of multiple network, complex network structure with multi task allocation, in task processing will produce parallel oscillation, the oscillation in complex networks is an important fac-tor to improve the inhibition of complex network parallel processing. Parallel oscillation by using the traditional method of singular value decomposition to reduce the dimensionality of the matching algorithm, the memory requirement and the time loss a lot of in the face of large-scale complex task is solved. This paper proposes a parallel oscillation suppression algo-rithm based on dual decomposition of hidden groups, first carries on the complex network of multi dimension business paral-lel processing model design, the index system of multi dimension business parameters of complex network parallel process-ing model, implicit parallel feature matching method to realize the double feature matching parallel processing. Simulation results show that, the parallel feature matching by using the algorithm of hidden groups of complex network, to realize the parallel processing and serial processing, the least square singular value decomposition time cost and space cost is reduced greatly, improve the speedup of 2 times, effectively inhibit the network oscillations. Algorithm in the complex network of multi task parallel processing in the number, it has the superior performance of delay and energy efficiency.