A Study of Costs and Benefits of Content Sharing in Personal Cloud Storage

Abstract

Personal Cloud Storage (PCS) is a very popular Internet service. In addition to backup, PCS allows content sharing among multiple devices, which is a valuable functionality for many users. Yet maintaining a PCS service incurs both storage and bandwidth costs to service providers, as the update and successive downloads of shared files may generate extra traffic on the PCS cloud servers. This becomes particularly worrisome as a large fraction of users (e.g., over 90%) does not pay for the service, joining a free version with limited storage capacity but typically without content sharing restrictions. Thus, a natural concern arises on the costs and benefits of such service, for both provider and users. In this paper, we propose a model to analyze the cost-benefit tradeoffs of content sharing in PCS services for both parties. Our model uses a macroeconomic concept, notably user surplus, to capture the satisfaction of different classes of users, as well as a cost saving function to capture the interest of the provider in reducing bandwidth costs. We use our model to evaluate two alternative policies to the content sharing architecture in use in existing PCS services, searching for scenarios in which both parties have benefits. The proposed policies rely on incentives given to users in exchange of their participation to offload shared content from cloud servers. Our investigation, based on analytical modeling and data-driven experiments, shows that the incentive leads to greater satisfaction to both parties, and that the alternative policies can reach scenarios which benefit both provider and users, with reductions in provider’s bandwidth costs by 20% and increases in user satisfaction from 51 to 82% under reasonable model assumptions.

Appendix 1: Components of the Cost of Service

In this section, we show how to estimate the components of the cost of PCS services for paying users, as they are used to compute the user’s surplus. It is worth noting that most PCS services do not express the components of the service price explicitly, as the bundling price is adopted, i.e., most services sell packages of bytes. Our goal then is to show that optimal estimates for these components can be obtained based on the bundling price (packages of bytes).

We can estimate the fixed component (f) and per byte component (p) of the cost of service for paying users (Eq. 5), i.e., the two-part tariff, by adopting the approach proposed in [25]. We first compute the unit price per byte u(x) of the bundling price for each byte x from 1 to X (the largest $x_{ij}$) and equals them to the two-part tariff unit prices by:

$$\begin{aligned} ~ u(x) = \frac{f + p \times x}{x} = \frac{f}{x} + p. \end{aligned}$$

(12)

Next, we estimate the values of components per byte ($\hat{p}$) and fixed ($\hat{f}$) that minimize the distance between two-part tariff and bundling unit prices by ordinary least squares:

$$\begin{aligned} ~ Q = \sum _{x=1}^{X} (\frac{\hat{f}}{x} + \hat{p} - u(x))^2. \end{aligned}$$

(13)

Figure 13 shows unit prices of the bundling (black curve) and estimates of the two-part tariff (red curve) for two of the largest PCS providers at the time of this writing: Dropbox offers the packages Plus and Professional with volumes 1000 GB and 2000 GB and prices $\$ 9.99$ and $\$ 19.99$ respectively, whereas, Google Drive offers two packages with capacities 100GB and 1000GB and prices $\$ 6.99$ and $\$ 34.99$ respectively.

Fig. 13

Unit prices of Dropbox (a) and GDrive (b) for paying users (y axis log)

We note the estimated two-part tariff ($\hat{p}$ and $\hat{f}$) follows the main trends of the bundling unit prices, as expected for optimum values that minimize the distance between those unit prices.