Abstract. The number of mobile devices (e.g., smartphones, tablets, wearable devices) is rapidly growing. In line with this trend, a massive amount of mobile videos with metadata (e.g., geospatial properties), which are captured using the sensors available on these devices, are being collected. Clearly, a computing infrastructure is needed to store and manage this ever-growing large-scale video dataset with its structured data. Meanwhile, cloud computing service providers such as Amazon, Google and Microsoft allow users to lease computing resources with varying combinations of computing resources such as disk, network and CPU capacities. To effectively use these emerging cloud platforms in support of mobile video applications, the application workflow and resources required at each stage must be clearly defined. In this paper, we deploy a mobile video application (dubbed MediaQ), which manages a large amount of user-generated mobile videos, to Amazon EC2. We define a typical video upload workflow consisting of three phases: 1) video transmission and archival, 2) metadata insertion to database, and 3) video transcoding. While this workflow has a heterogeneous load profile, we introduce a single metric, frames-persecond, for video upload benchmarking and evaluation purposes on various cloud server types. This single metric enables us to quantitatively compare main system resources (disk, CPU, and network) with each other towards selecting the right server types on cloud infrastructure for this workflow.
IntroductionWith the recent advances in video technologies and mobile devices (e.g., smartphones, tablets, wearable devices), massive amounts of user generated mobile videos are being collected and stored. According to Cisco's forecast [7], there will be over 10 billion mobile devices by 2018 and 54% of them will be smart devices, up from 21% in 2013. Accordingly, mobile video will increase 14-fold between 2013 and 2018, accounting for 69% of total mobile data traffic by the end of the forecasted period. Clearly, this vast amount of data brings a major scalability problem in any computing infrastructure. On the other hand, cloud computing provides flexible resource arrangements that can instantaneously scale up and down to accommodate varying workloads. It is projected that the total economic impact of cloud technology could be $1.7 trillion to $6.2 trillion annually in 2025 [8]. Thus, the large IT service providers such as Amazon, Google, and Microsoft, are ramping up cloud infrastructures.One key question is how to evaluate the performance of mobile video applications on these cloud infrastructures and select the appropriate set of resources for a given application. Suppose a mobile user wants to upload a video to a cloud server along with its metadata (e.g., geospatial properties of video such as camera location and viewing direction), which are captured and extracted using the sensors embedded on the mobile devices. Note that this kind of geospatial metadata enables advanced data management, especially in very...