Last week I lost the hard drive in my Macbook Air after about 3 months of it playing the Little Drummer Boy in that hard drive clicking language only terminally ill HDs can speak. It is remarkable that in over 15 years of owning Apple laptops, this is the first drive I've lost, though perhaps more indicative of the fact that it is just a bad idea to expect an iPod drive to be able to do the work of its bigger brethren.
As I attempted to reconstitute the machine, it occurred to me that we are still in the stone ages of virtualizing consumer storage so that people can really treat their small computing devices as active caches instead of as precious sources of data— which is a strange notion given how crowded the space of consumer backup sector is: from small but wonderful services like Dropbox, through at-scale startups like Carbonite, all the way to platform vendor services like Upline and MobileMe, it seems that no one has gotten it quite right in that there is no one place where you can go to seamlessly return to a known working state of your trashed device.
I use a 3-tier system to keep a few laptops in relative sync/backed up, and I can't help but think that this layered approach might actually be the right solution (for now), if only the players at each layer would focus on working well at their layer instead of covering the whole space. Here is how it works:
First, there is the data that is tightly coupled to the application but which needs to instantly replicate across all of my machines and other portable devices, specifically calendar, addressbook, and to-do info. From a shoveling bits perspective, this is really small and therefore easy to copy over even the slowest of networks. The challenge is in sync (and it is remarkable that there is still a challenge here given how many people have beaten their heads against this for the past 2 deacdes), and in making sure that as soon as I make a change on one client, it is replicated to the server and then to all of the other devices. Right now I use a combination of MobileMe and Exchange for this— and it is truly remarkable how poor each of the solutions is— a suckage that is only multiplied when one tries to use them together.
Second, there is data that is loosely coupled to specific applications but in need of quick replication across a number of machines. These are the Office docs, source files, graphics assets, etc. that I might be actively working with at any given moment. Not a huge amount of data, but enough that you have to be somewhat smart about replication to keep from constantly overtaxing the network. A good source control system (I use subversion) is a great model for how this tier of data replication should work, and in fact the solution I've come to use for non source control files— a startup called Dropbox— is just a prettied up source control repository with a set of great clients. Most of the startups in the "consumer storage" space play in this tier, though they tend to spread themselves too thin and (unlike Dropbox) forget about the importance of having a really fast and reliable client.
Finally, there is the "everything else" which should represent everything from / on the machine's drive. I'm talking about applications, application support data, configuration files, etc. On the drive I lost this was about 40GB of tweaked OS & apps that I had spent close to a year customizing. The theory on this tier is that in the near term you won't see enough pipe to be able to back it up offsite so a local solution is necessary. I use the Apple Time Capsule product, which is basically a 1TB drive hooked into a wireless router and a client app that performs full disk backups on a schedule. It works ok— not great— mostly because the client is not good at understanding that laptops are opened and closed a lot and that it needs to be smart about pausing/resuming the backups accordingly. Also, the restore process doesn't overwrite the Apple-installed applications which is deadly when you've done thinks to tweak those, or say when your tightly bound application metadata has been changed by a later version of the application than what the reinstalled operating system has.
Eventually (maybe a decade from now), we'll all go to a system where everything is replicated instantly to some cloud provider— maybe even with a VM instance of the running machine being continuously streamed up for to-the-moment restoring of state. But until we get to this Star Trek universe if I were in this space, I might think about these three layers, decide where it is that I want to play, and focus maniacally of the details that still make each of them suck so much.