面向海量自体的检测器反向生成算法

Reverse Detector Generation Algorithm for Massive Self Set

检测器生成算法是影响人工免疫系统性能的重要因素之一，在大数据环境下由于自体数量的庞大使得现有检测器生成算法无法在有限时间内构建出成熟检测器集．在前期使用MapReduce模型构建分布式检测器生成系统的基础上，分析影响算法效率的主要因素；设计了MapReverse Reduce模型构建检测器反向生成算法；通过Reverse阶段反转Map阶段的检查结果并将非法检测器键值对发送给Reduce阶段进行成熟检测器筛选，提高海量自体时人工免疫系统生成检测器的效率；最后在Hadoop集群中分别使用MapReduce模型和MapReverseReduce模型实现检测器生成算法的原型系统，并使用CERT synthethic sendmail data数据集进行测试与分析，验证了使用MapReverseReduce模型生成检测器的时间开销只有使用MapReduce模型时的5．22％-19．07％，并在自体数量不断增加时保持算法时间开销的稳定．

Detector generation algorithm is important for artificial immune system. When implementing the artificial immune algorithm in big data system,the huge self will lead to the large time and space cost. In this paper,we analyze of the major factors affecting the efficiency of the distributed detector generation algorithm base on MapReduce model. Then we present the MapReverseReduce model and use it to design the reverse detector generation algorithm. We use Reverse stage to reverse the result of initial detector inspection from Map stage, then send the key-value pair of illegal detector to select the mature detector in Reduce stage. The Map, Reverse and Reduce can work in parallel. It can decrease the inspection between initial detectors and self,reduce the number of key-value pair that Reduce stage should deal with and the communication between Map and Reduce stage. The efficiency of detector generation can be improved obviously. Finally, we realize the prototype of distributed detector generation algorithm using MapReduce model and reverse detector generation algorithm respectively. Using the CERT synthethic sendmail data set to test the time overhead of detector generation, and different stage in reverse detector generation algorithm. The results show that the time overhead of reverse detector generation algorithm is of 5.22 to 19.07 percent of using MapReduce model, and time overhead can been maintained stability with the number of self increasing.