The first question Mark and I agreed on was to go over a quick history of configuration management. I know of no one with Mark’s long-standing commitment and presence in the CM field. It seemed like a natural question to me and one which everyone with a bit of background in the field might find interesting. I mean, the guy has seen the whole movement arrive with all of its autonomative iterations (immunological, descriptive, declarative, etc).
The “treat your servers like cattle, not like pets” (similar to the strongly resonating “servers should be cattle, not snowflakes”) is a maxim I first heard from DevOps heavyweight Damon Edwards. It continues to enjoy some popularity on twitter as it so succinctly captures one of the problems that DevOps takes to task: stop keyboarding stuff you should have automated; stop treating servers like cherished curata; snowflakes belong on the ski slopes, not in the data center.
And all this history? Just goes to show the world isn’t resting with today’s CM, regardless of what you may hear in trade show aisles from the vendors (or what I heard attending LISA12…).
Tomorrow’s CM will be recognizable to us today, but probably only in some bare and basic ways. As an analogy to what I mean, have you ever tried to start or drive a Model-T? (We’ll be exploring that topic more later on in the excerpts with Mark.)
Mark Burgess: So there’s a human part of the question and a machine part of the question.
In terms of the technology clearly, there’s been a revival of interest in the idea of configuring systems with automatic software, which is something that I got into in the early 90s actually and that has been through a number of transformations.
And then there’s also [ . . . ] a look at the problem of infrastructure management and configuration management and how their attitudes have changed, too.
And I think both of them are kind of interesting questions.
So on the human side I think, throughout the 90s, people gradually got out of the habit of managing machines by hand because we started the 90s with just a few machines — tens, perhaps, and then hundreds, maybe thousands towards the end of the 90s, but we start out by managing machines like beloved pets and stroking them with our bare hands and typing away on the keyboards and they became our friends.
Tom Parish: …yeah, we named them.
Mark Burgess: We named them, exactly. And that builds a relationship between humans and machines in a very particular way.
Tom Parish: Yes.
Mark Burgess: And I think as the 90s grew and as the number of machines grew it became maybe almost a little sad but we weren’t able to scale up that closeness to large numbers of machines. [ … ] So — and then we got things like the network shells, you get gold discs and things that are pushing out golden images and all of these kinds of mass crowd control systems for managing.
But towards the end of the 90s I guess that’s when I’d put in a lot a work in CFEngine to try to get actually, if I jump to ’98, I read a paper called “Computer Immunology,” where the idea was to think of the computers managing themselves. You know, managing themselves from within, like our bodies sort of look after themselves from within.
And that kind of changed, for a few people at least, there is the “black belt” segments, they liked this idea of automating the configuration of systems and doing it from within so that they could work in a hands-free kind of a way and then build a relationship with policy or a configuration design, rather than the individual machines themselves.
So that’s kind of the human aspects of the 90s.
And then of course the web came along and messed everything up, if you asked me. Those guys, I mean, come on. And then of course with the key thing there was make as many machines as you can and fast as possible.
Tom Parish: Right. Of, course.
Mark Burgess: And of course that diluted the black belt assembly so a lot of new blood coming in, maybe not as experienced, and then the focus shifted back towards building machines and setting them up.
Because I think what kind of happens through the 90s was that we looked more about maintaining the health of systems, hence the title of ‘computer immunology’.
It wasn’t just enough to set them up, you wanted to keep them running. And you remember, it was like comparing how long my machine had been up, compared to everyone else’s …
Tom Parish: … certain pride in that.
Mark Burgess: Exactly. And that was “200 days”, “300 days”, “More than a year since the last reboot.”
Nowadays it’s like we don’t care, we reboot the machines willy-nilly… so that’s the human story, and then the technology has kind of tracked that a little bit.
So CFEngine 2, which was kind of the first immunological approach just to management, reached its peak, maybe around 2000 time, and then as the web came in, you saw these other tools […] come in that were built much more around provisioning systems.
And then we’ve come almost back towards seeing management as important, I think, the cloud’s been through an introductory period and we’ve had this flirtation with building things […] now I think actually people are starting to come back towards stability and maintenance and some of that pride in the systems that perhaps waned a little bit.