Check out the new USENIX Web site. next up previous
Next: 5.2 Limitations Up: 5 Comparison with existing Previous: 5 Comparison with existing


5.1 Cost

We did a survey of Internet backup sites on October 28, 2002. Figure 8 shows the average and minimum monthly fees for various amounts of storage for the 15 sites (out of 28 surveyed) that list prices on the web for given amounts of data to be stored (as opposed to the amount to be backed up, which differs due to assumptions about compression and how much data actually needs to be backed up). The cheapest marginal cost found was US $7.20 per gigabyte.

Figure 8: Monthly fees in US dollars required by existing Internet Backup services to store a given amount of data; excludes one-time startup fees but includes discounts for annual contract.
\begin{figure}\begin{tabular}{l\vert r\vert r\vert r\vert r\vert}
space & 100~MB...
...min & \$4.50 & \$12.71 & \$72.21 & \$720.00 \\
\hline
\end{tabular}\end{figure}

In estimating the cost of using our scheme, we assume that the user's computer hardware, base power, and any needed bandwidth are already paid for by existing uses. This seems reasonable for a home PC that is already on enough for reasonable predictable uptime and that has an Internet connection with flat-rate bandwidth pricing. We furthermore assume no cost for the software or central-server operation based on an open-source model and the extremely low overhead on the central server per participating computer; if a commercial model of software development was used instead, a small one-time fee for the software might be also be required. Under these assumptions, the cost of our scheme is determined by the marginal cost of storage for a PC and the marginal cost of the extra power needed to operate a disk drive to answer challenges and backup data.

Based on the cost of a 60 GB internal IDE hard drive as of October 2002--US $75 according to www.pricewatch.com--depreciated over 2 years, we conservatively estimate the marginal cost of storage at no more than 5.2 US cents per gigabyte per month. The cost of the extra power required is much harder to estimate. One conservative approach is to assume that the disk is turned on 1.5 extra hours per day per gigabyte to be backed up, on the assumption that the average backup takes 25% of the time of a full one (2 hours/GB for the prototype in each direction) and that challenges take half an hour of disk time total per day. Desktop disk drives appear to consume about 10 watts extra power when active [11] so at a conservative electricity cost of 15 cents per kilowatt hour, our scheme should use 7.5 US cents of power per GB backed up per month.

If we assume 100% storage overhead for redundancy (e.g., $k{=}m$) and room for 2 full snapshots on the logical disk (a factor of 2.2 should suffice), we will need to trade 4.4 GB of local disk in order to back up 1 GB of data, resulting in our scheme costing no more than 26 US cents per gigabyte per month. A less conservative estimate (3 years, 50% overhead) gives a figure of 18.6 US cents/GB/month.

Thus, our scheme appears to be 30 to 100 times cheaper than existing Internet backup services. We do not fully understand why this is, but believe it largely stems from our scheme's lack of administrative costs and use of marginal resources (e.g., most of the resources we use are already paid for by other uses). A small part of the difference may be due to limitations of our scheme as compared to existing Internet backup services (see below).

Traditional backup methods can be comparable in cost to our scheme, but are inconvenient for users: if a home-computer user weekly writes a snapshot to 700 MB CD-Rs (6 US cents each in quantities of 100) then takes them to work and leaves them there, he will incur a cost of 35.1 US cents/GB/month ignoring the cost of the CD burner.


next up previous
Next: 5.2 Limitations Up: 5 Comparison with existing Previous: 5 Comparison with existing
Mark Lillibridge 2003-04-07