Factory Joe

I had an interesting discussion with Freeman Murray at SHDH last night about offline apps for use in remote areas where network latency can sometimes stretch from days into weeks for access to a “network”.

In such circumstances, you’ll have folks driving around on mopeds or busses with wifi antennae, USB drives or CD-ROMs delivering email and providing a means to getting “online”, admittedly asynchronously. Thought 28.8 was slow? Try email by bike messenger. It’s only one step above message delivery a la Paul Revere.

But in some cases, this is the best they’ve got and you can’t expect Google to trot around the world setting up mesh wifi networks for these folks (not in the near future anyway). And while some folks are working on this project and it is getting better, nevertheless, we must constantly be aware of and design for a rich offline experience.

This is especially close to me, considering I’m writing this on BART without connectivity as I head to the airport. Yes, even in industrialized areas connectivity is still not guaranteed (let alone free).

So an idea I’ve been playing with for some time is how Flock can support offline browsing and interaction, beyond pulling simple content from your cache or downloaded feeds. In the old model of the static web, the permalink model made sense and was perfectly useful — indeed, it’s still nice to be able to pull up that bar’s website when you’re wandering around Paris at 9pm, an hour late for the meetup that you helped coordinate and you’ve got no idea where it is except for the micromap saved in your browser’s cache.

But the problem is that we’ve divorced the data from the interaction layer. It’s like having your Halo saved games without having the game engine to play them. Boy, that’s a whole truckload of fun.

So anyway, Freeman and I were discussing how Flock could help with this problem. One idea that has some legs, I think, combines data delivery in microformatted XHTML pages (you’re caching it already, might as well make the data semantic and thus useful) and then allow basic interactivity with that data through extensions that are designed for both online and offline use. Thus, in the event that you’re offline, well, instead of pulling unsuccessfully from the server, it would pull from your local cache and allow you to do certain basic things, queueing your actions to be performed when you have connectivity again.

Indeed you could even use a P2P architecture for this, sharing encrypted offline action-scripts across a mesh network. When any one of those nodes reconnects, it could upload those instructions to the final destination where they’d be executed in batch. This would have the added benefit of spreading the need for connectivity across many nodes, instead of just one (like a USB drive which could get lost, stolen, confiscated or otherwise compromised). Should the actions have already been performed when the “message in a bottle” arrives, the commands would be simply ignored. There are technical details in here that are beyond my comprehension, but be that as it may, the idea is promising.

So back to offline extensions development… Freeman proposed an architecture in which Flock would ask the web app for a locally-run “servlet” that would provide similar offline interactivity when not connected. Autodiscovery would happen in the way that sites provide favicons or perhaps through a rel=”offline” link. The question though, is whether the user would need to take explicit action to install the servlet extension outright, or if, by visiting a web app (like Gmail or Basecamp), you’re implicitly expressing your interest in using the functionality of the app, whether on or offline.

I think the latter reflects a reality that I want. Installing apps on the desktop doesn’t make sense anymore, what with all my data being hosted in the cloud. So being to access and manipulate my data when I’m offline is going to become a requirement that the browser can tackle. I mean, this is the problem with solar power. You need to use your car whether it’s sunny or not, and so we’ve got rechargeable batteries, right?

Well, I think you get the point. Here’s the high-level description of the proposed solution (which I may or may not have communicated effectively):

• microformatted XHTML cached as your local datastore
• locally-run servlets that offer data interactivity simulating the remote data store
• offline actions supported in extensions
• queueing mechanism in the browser to synchronize via XML-RPC or p2p encrypted action file when connectivity is available

I probably got a bunch of the technical details wrong, but whaddya think of the general concept? Does this already exist in some other form somewhere?