Deliver the content over multiple surrogates: A request routing model for high bandwidth requests

Abstract

Abstract The Content Delivery Networks (CDNs) deploy the surrogate servers at the network edge to speed up the request routing procedure. But, the routing procedure of high bandwidth requests increases the load of one server suddenly. This overloading also rises the queue waiting time of the corresponding surrogate server. The higher queue waiting time also increases the response time and drop rate in that server. As a solution to these problems, we propose a novel request routing procedure which focuses to deliver the content over multiple surrogates. In this request routing procedure, we decompose the high bandwidth content requests of clients to more than one surrogate server. Accordingly, we first combine the client requests at the origin server. Then, these requests are kept in the queue of the origin server and received basic content to find the congestion status. We also estimate a bound for the origin server and compare it with the congestion status to determine the high bandwidth requests. Finally, we decompose these requests to more than one surrogate server by adjusting the split size according to the upcoming queue load and waiting times. We utilize ARMA model by combining the autoregressive and moving average models to estimate the next values of these parameters. Therefore, the congestion status of the origin and surrogate servers are used for the determination of high bandwidth content requests and split size adjustments of whole content, respectively. Also, to determine the queue load and waiting time parameters, we model the origin and surrogates according to the G/G/1 queuing system. All request routing procedures are executed by the controller through the superposition, queuing, comparison, splitting, estimation, adjustment, and combination modules. According to the simulation results, the response time and request drops are decreased by 36% and 45% compared to the conventional approaches, respectively. 1