Fifteenth anniversary of the Halloween documents

Fifteen years ago, the "", were released by an unauthorized tipster, and published to the internet by Eric S. Raymond.These were a set of confidential memoranda within Microsoft describing the threat of Linux.If you recall, Microsoft had an overwhelming monopoly among computer operating systems at the time.What made the documents so controversial was that the internal deliberations at Microsoft were completely out of step from their public statements about Linux.Microsoft had dismissed and ridiculed Linux in public but internally it was agreed that such a strategy would not work in the long term.

"OSS is long-term credible ... FUD tactics can not be used to combat it." This quote captures several important aspects of the memos.First, there was a realization that open-source software was high quality, and that the competition lay in the open-source community rather than a particular software release.At the time, there were hackles and smirks about the fact that underhanded marketing approaches like "FUD" (or fear, uncertainty, and doubt) were discussed so casually.




This work also led to several academic views of the topic.A landmark example is the case study written by Lerner and Tirole, "The Simple Economics of Open-Source Software." [].This paper reviewed the history of open source software, described the context of the programming culture, and described four examples of open-source software: Apache, Perl, Linux, and Sendmail.



At the time, I was fascinated by the network effects of open source software and the prospect of characterizing its dynamics in an epidemiological sense.For instance, what are the dynamics that govern the spread of a release of open-source software?In this context, the software growth and adoption could be modeled as if it were a bacterial infection.Moreover, this is loosely connected to the notions aboutand modeling of nonlinear systems which were prevalent in the late 90s.In particular, software can be replicated and forked -- which correspond to "reproduction" and "mutation" in a population genetics sense.The Artificial Life people make simulations which had these characteristics and then assert that the entities being simulated are "alive."In the open-source software case, we don't need a simulation -- we just need to statistically track the software developers!



One especially interesting item in the Halloween documents referred to the effect of the software license has on code forking [].In particular, they made the observation that the permissiveness of the BSD license encouraged code forking, but the viral GNU license discouraged it.In a dynamical system composed of equations, this would correspond to certain notions of stability.



However, I found the modeling of the software development process a challenge.It is difficult to convincingly model this behavior -- though it is probably more tractable now in the era of big data!Instead, I focused on the economics impact of open-source software [].The is more relevant literature here from which to justify a model, particularly in the simplified case of no switching costs.In this case, the model is composed of simple agents that represent the companies (including open source), the vendors, and the users.



It turns out that although users may have complex reasons for making a purchase, but the model approximates this pretty straightforwardly.The users purchase the most convenient software within their budget and the order in which they consider companies is a stochastic function which is weighted by the advertising budget from a given company and the market share of a given company.There is presumed to be no variable costs, and the companies act as profit maximizers -- with advertising being proportional to profit.In the case that the price sensitivity of users goes all the way down to zero, the pure software companies cannot make money and eventually go away, which is predicted as follows in the below plot:



There is a dominant company and a secondary company in equilibrium when an open-source product is introduced at time=0.The open-source product takes out the 2nd company almost immediately, but there is a longer period in which the dominant company makes various adjustments (mainly increases in price) until they cannot compete.

However, the situation changes when the users have a price sensitivity floor.This price floor might be used a a proxy for the usage cost, below which price differences do not cause users to make decisions.In this case, the model predicts that it is possible for open-source products to co-exist with commercial products, as depicted below:

Here, the case is that the 2ndary company is affected right away and loses its revenues (along with several other players that can never get a significant portion of the market with or without open-source).However, in this case the monopolist makes some small adjustments and is never affected.

The reality is more complicated, but the modeling does provide some useful insights which could be transferable. For example, there is an open question right now regarding the dichotomy between soaring costs of higher education and the availabilty of free on-line courses (MOOCs).Apple just released the latest version of their proprietary operating system for free.Wouldn't it be interesting if fifteen years from now, Harvard offered an undergraduate off-campus education, with a different but equivalent certification, for free as well?
Full Post