Abstract—Media streaming applications have recently attracted a large number of users in the Internet. With the advent of these bandwidth-intensive applications, it is economically inefficient to provide streaming distribution with guaranteed QoS relying only on central resources at a media content provider. Cloud computing offers an elastic infrastructure that media content providers (e.g., Video on Demand (VoD) providers) can use to obtain streaming resources that match the demand. Media content providers are charged for the amount of resources allocated (reserved) in the cloud. Most of the existing cloud providers employ a pricing model for the reserved resources that is based on non-linear time-discount tariffs (e.g., Amazon CloudFront and Amazon EC2). Such a pricing scheme offers discount rates depending non-linearly on the period of time during which the resources are reserved in the cloud. In this case, an open problem is to decide on both the right amount of resources reserved in the cloud, and their reservation time such that the financial cost on the media content provider is minimized. We propose a simple—easy to implement—algorithm for resource reservation that maximally exploits discounted rates offered in the tariffs, while ensuring that sufficient resources are reserved in the cloud. Based on the prediction of demand for streaming capacity, our algorithm is carefully designed to reduce the risk of making wrong resource allocation decisions. The results of our numerical evaluations and simulations show that the proposed algorithm significantly reduces the monetary cost of resource allocations in the cloud as compared to other conventional schemes.
INTRODUCTION
Media streaming applications have recently attracted large number of users in the Internet. In 2010, the number of video streams served increased 38.8 percent to 24.92 billion as compared to 2009 [1]. This huge demand creates a burden on centralized data centers at media content providers such as Video-on-Demand (VoD) providers to sustain the required QoS guarantees [2]. The problem becomes more critical with the increasing demand for higher bit rates required for the growing number of higherdefinition video quality desired by consumers. In this paper, we explore new approaches that mitigate the cost of streaming distribution on media content providers using cloud computing.
A media content provider needs to equip its data-center with over-provisioned (excessive) amount of resources in order to meet the strict QoS requirements of streaming traffic. Since it is possible to anticipate the size of usage peaks for streaming capacity in a daily, weekly, monthly, and yearly basis, a media content provider can make long term investments in infrastructure (e.g., bandwidth and computing capacities) to target the expected usage peak. However, this causes economic inefficiency problems in view of flashcrowd events. Since data-centers of a media content provider are equipped with resources that target the peak expected demand, most servers in a typical data-center of a media content provider are only used at about 30 percent of their capacity [3]. Hence, a huge amount of capacity at the servers will be idle most of the time, which is highly wasteful and inefficient. Cloud computing creates the possibility for media content providers to convert the upfront infrastructure investment to operating expenses charged by cloud providers (e. g., Netflix moved its streaming servers to Amazon Web Services (AWS) [4], [5]). Instead of buying over-provisioned servers and building private data-centres, media content providers can use computing and bandwidth resources of cloud service providers. Hence, a media content provider can be viewed as a re-seller of cloud resources, where it pays the cloud service provider for the streaming resources (bandwidth) served from the cloud directly to clients of the media content provider. This paradigm reduces the expenses of media content providers in terms of purchase and maintenance of over-provisioned resources at their data-centres.
In the cloud, the amount of allocated resources can be changed adaptively at a fine granularity, which is commonly referred to as auto-scaling. The auto-scaling ability of the cloud enhances resource utilization by matching the supply with the demand. So far, CPU and memory are the common resources offered by the cloud providers (e.g., Amazon EC2 [6]). However, recently, streaming resources (bandwidth) have become a feature offered by many cloud providers to users with intensive bandwidth demand (e.g.,
Amazon CloudFront and Octoshape) [5], [7], [8], [9].
The delay sensitive nature of media streaming traffic poses unique challenges due to the need for guaranteed throughput (i.e., download rate no smaller than the video playback rate) in order to enable users to smoothly watch video content on-line. Hence, the media content provider needs to allocate streaming resources in the cloud such that the demand for streaming capacity can be sustained at any instant of time.
The common type of resource provisioning plan that is offered by cloud providers is referred to as on-demand plan. This plan allows the media content provider to purchase resources upon needed. The pricing model that cloud providers employ for the on-demand plan is the pay-peruse. Another type of streaming resource provisioning plans that is offered by many cloud providers is based on resource reservation. With the reservation plan, the media content provider allocates (reserves) resources in advance and pricing is charged before the resources are utilized (upon receiving the request by the cloud provider, i.e., prepaid resources). The reserved streaming resources are basically the bandwidth (streaming data-rate) at which the cloud provider guarantees to deliver to clients of the media content provider (content viewers) according to the required QoS.
In general, the prices (tariffs) of the reservation plan are cheaper than those of the on-demand plan (i.e., time discount rates are only offered to the reserved (prepaid) resources). We consider a pricing model for resource reservation in the cloud that is based on non-linear time-discount tariffs. In such a pricing scheme, the cloud service provider offers higher discount rates to the resources reserved in the cloud for longer times. Such a pricing scheme enables a cloud service provider to better utilize its abundantly available resources because it encourages consumers to reserve resources in the cloud for longer times.
This pricing scheme is currently being used by many cloud providers [10]. See for example the pricing of virtual machines (VM) in the reservation phase defined by Amazon EC2 in February 2010. In this case, an open problem is to decide on both the optimum amount of resources reserved in the cloud (i.e., the prepaid allocated resources), and the optimum period of time during which those resources are reserved such that the monetary cost on the media content provider is minimized. In order for a media content provider to address this problem, prediction of future demand for streaming capacity is required to help with the resource reservation planning. Many methods have been proposed in prior works to predict the demand for streaming capacity [11], [12], [13], [14].
Our main contribution in this paper is a practical—easy to implement—Prediction-Based Resource Allocation algorithm (PBRA) that minimizes the monetary cost of resource reservation in the cloud by maximally exploiting discounted rates offered in the tariffs, while ensuring that sufficient resources are reserved in the cloud with some level of confidence in probabilistic sense. We first describe the system model. We formulate the problem based on the prediction of future demand for streaming capacity (Section 3). We then describe the design of our proposed algorithm for solving the problem (Section 4).
The results of our numerical evaluations and simulations show that the proposed algorithms significantly reduce the monetary cost of resource allocations in the cloud as compared to other conventional schemes.