Scheduling policy analysis of cloud video service

As a leader in providing video service in US, Comcast provides cloud video service to Americans. Currently, the primary legal interpretation approving cloud video recording in cable-vision decision relies on individual copies for each video recorded. This means every recorded video in the cloud needs its own copy in the data center. Categorizing users by the estimated similar behavior is proposed manage the 20 billions videos available to around 50 millions subscribers. In this system, the subscribers with similar behavior are grouped in one cluster. Cloud recording cable video and IP video service requests are asynchronous and stored in two queues. The cluster goes to sleep if neither queue has requests. The cluster wakes up when a new request arrives. The cluster takes heterogeneous service calls: recording cable video and IP video. In this paper, a 2-class Markov Geo/G/1/K vacation model is presented to analyze the scheduling policy of processing two queues.