Friday, August 6, 2010

Nail 2.0

I live in Berkeley. Most of the houses here pre-date me. Almost all are built with wooden frames, which means almost all were built using nails. Imagine, if you haven't already, that the Naitional Nail Authority has issued Nail 2.0. Immediately, however, a bug in the Nail 2.0 implementation causes the underlying Nail 1.0 implementation to AB-END and fail to build. So immediately, of course, all Nail 1.0 and later spec nails cease to exist. So 95% of the houses in Berkely immediately fall over, and the town looks like the last little pig was smart after all, until owning a nice brick home in a town of wooden houses all broken up with gas lines open all over the place starts to seem like a really bad idea.

This kind of Scorched Earth vision was all the rage in the last part of 1999. I was working in Financial Services doing portfolio analytics server-side in Java with a legacy back end of millions of lines of FORTRAN code. So we didn't sweat the Y2K bug at all -- financial analytics really have to be able to go way into the future for Bonds and what-ifs, so the system had always been written to handle Y2K dates. But I remember thinking that if the EastBayMUD servers, which I had toured, and were highly COBOLized, were to crash, then for a few days we might not have water. Though more likely was just an EBMUD water bill for a century's worth of water.

So why don't houses fall down when they issue a new nail? Because objects in the real world are manufactured, and then installed and left there with no dependencies. Having a build system is a luxury deployed by just about every software shop, and so is having developers who use version control, test suites, etc. Yet, when we build software, we don't package everything up and ship appliances. We ship tiny bits of code that depend on lots of other code. And our bits of code were made by the interaction of thousands of other littler bits of code. And we send all our little pieces, with all their dependencies on pieces that we don't ship, out in their little boats. Sorry, lost the house metaphor for a boat reference. But I like the imagery of it.

In the dot-com binge, my brother and I started and designed what is now called Cloud Computing. We designed, coded, networked, prototyped, and documented most of, say, what you can get from Amazon EC2. My brother worked on selling it to Dell, Compaq, and some large ISPs. We got really close, but we got beaten to market by several competitors.

I knew going into it that the ideas I had were the result of being in a computer geek zietgiest that probably included about 2 million really sharp minds. I knew there would be lots of others with identical startups, because appliances had already busted out, and linx had GUI terminal servers, so the concept of provisioning a blade at boot time with a networked version-controlled repository of boot images and large data mounts seemed like the way to go since it was clear that big iron would make some kind of comeback. I was also working then on designing web services so that all our virtualized boxes could have very simple duties and just network their messages back and forth. But like I said, that's no boast--lots of people were all scratching their heads and thinking about the future and how it might be manageable with all these dang Linux boxes stacked in the server room and all these dang software pieces.

So now we have Cloud Computing and we have Web Services. But we are still missing some of the things on my wish list.

Real packaging. You get the immutable bits that will work. All of them.
You can still have a build system for assembling and testing these beasts. But don't expect users to have a build system that can handle your kruft.

Simple communication.
All text communication.
Bi-directional communication supported, and one-way messaging is preferred.
To implement this today, I'm thinking: XML messaging. XML file formats. Use schemae and attach them to the document. These all seem viable and established now.

Real interfaces. Typed, inheriting, globally known, versioned, assembled interfaces. (Possible with things like XML messaging and JSON. But also perhaps with sending objects around.)

Come up with some kind of object that can be sent around the network and stored and versioned and transformed, round-trip, to text and back. Java does this, and it plays well with real object databases. I suspect Javascript could do this well, too, as my brilliant friend Brian hipped me to, but he said he detests blog technology, so I guess I shouldn't link to him. Anyway, he says it's totally doable, and folks are doing it.

But I haven't had too much time to see what is up in the javascript-as-object-nirvana world, since I am happy as a clam since I got the schema for the objects coded in Java and stored in db40 and accessed with the rudiments of Retro (my command-line framework) working on my laptop. Java objects. Simple. Not "Value Objects," but POJOs. Properties, collections, relationships--all store nicely in db40. And I'm not doing any kruft. No annotations, no JAXB, no mapping file, no injection, no nothing.

I'll put up the source very soon, and design docs. For now, please read the other posts in this blog to see where I'm hoping to go.