Research On Algorithms For Service Replication During Network Flash Crowd

Flash crowd and DDoS are two of threats to the modern internet and its hosting services. They may keep server busy and cause network congestion, which degrade the QoS for client subsequency. The traditional strategies like Cluster and CDN could not handle Flash Crowd and DDoS efficiently since:(1) Falsh crowd and DDoS only occur by accident and the request pattern for a service provider is kept in a norm level. So the overprovisioning for a service is not uneconomical; (2) The behavior of Flash Crowd is identified by rapid rise in traffic and it is difficult to predict flash crowd and its peak. Therefor, CDN which has fixed capacity could not satisfiy the dynamic requirement needs in capacity.One of solution to solve Flash Crowd is to creat mutual-aid community. In this community, each peer could join and leave randomly and dynamical replication is trigged when flash crowd is detected. This paper focus on the algorithms for service replication during network flash crowd and my work are:1. We analyse the log of world cup web site http://www.france98.com and summarize the three characterizations of such web site. Then we provide a replication triggering model, in which a flash crowd event is identified when the change of request pattern grads excesses predefined thresh.2. We assume that all requests that excess system's capacity are discarded during flash crowd. We then classified all service into two types (TYPE_A and TYPE_B) and then adopt the discared requests during replication as our object. Three replication methods (unicasr, multicast and set multicast) are provided and four algorithms (RSA, EISA, IMMA and ASA) for set multicast are analyzed.3. In our experiment, we anlysized the releationships among parameters, objects and the performances of each algorithms. Next, we import the unused system capacity during replication to our object and use the scaled world cup web site's log as flash crowd as input of our simulation. Last we discuss the relationships between trigging thresh and object.