Yesterday I wrote about the need for a cron in the cloud and ended with the hope that someone builds a scalable way for periodic processes to run in much the same way that AppEngine allows for request-reponse interactions (turns out that it is already in the plan for AppEngine).

There are obvious reasons why a distributed scalable job manager is a great service, but today I want to pick up on why a common sandbox for all of the world's cron jobs might be a really interesting thing to develop.

My inspiration for my view of a cron in the cloud was really influenced by a recent talk I heard from my newest favorite author, Daniel Suarez (of Daemon fame) on "Bot Mediated Reality." Daniel is quickly coming to be regarded as someone who can pick out the faint outlines of the near future from existing technologies and in this talk he does not disappoint when he takes on the implications of the dangerous mix of society's ruthless quest for efficiency with automated programs deployed to take actions over a world whose interface is increasingly being standardized by web forms, APIs, and other such disinter-mediators of human judgment.

In his world, left unchecked, these bots will continue ruthlessly running on schedule taking first order actions that will cause the continued centralization of power in the hands of the few people capable of maintaining a semblance of control over the vast bot nets currently working for bad (DDOS style) and good (automated trading systems). Suarez's answer to avoiding the coming Judgement Day where the machines take over revolves around using richer interfaces to distinguish human from machine and forming committees of humans to control the types of bots we allow into a more secure and verifiable Internet (the "Dark Net" as he calls it).

But what if instead of doing that (or before getting there), someone like Google (or Amazon or anyone else who is credible in the cloud computing ecosystem) launches a platform for running bots that has built into it clear lines for distinguishing identity and (where appropriate) intent? I'm thinking of something a lot richer than "* * * * * run_me_every_minute.bin > email_errors—" something that encompasses identity, context, inter-job communication (with something a little richer than publish/subscribe), and maybe even some sort of a transaction system, just for starters.

Think about it. As the Web 2.0 lovefest melts down for real, one its true legacies is the legitimazation of the XML over HTTP API for inter application communication. At some conference last year I remember someone talking about how the inefficiency of the high volume polling that takes place to determine for instance whether a information source has been updated has a real and measurable impact on the carbon footprint of some of our largest web services. It might be worth doing Cloud Cron for just its green impact alone.

What I'm thinking is something like Second Life for processes with server farms partitioned and dedicated to all sorts of collaboration across processes. In Suarez's talk, he mentioned how deployed bots sometimes start working at cross purposes when one company acquires a competitor and the IT guys forget to turn off the competitive intelligence scrapers that have been looking at prices (or worse yet trading against) a former competitor.
I have no doubt this more reality than fiction even today, and no matter how brittle these jobs may be enough of them running unchecked may bite us in the butt someday soon.

So who wants to go out and build this cron for the cloud already?