The Cost (and Benefit) of Open Source

date:2006-04-24 23:59:05
category:Economics of Software

I’ve had this conversation more than once.

Me: “We can download something like POI to read the Excel™ files. Or I can spend months writing something.”

Them: “We have a policy against open source software.”

Me: “Do you use Apache ?”

Them: “That’s different.”

Me: “How so? Be specific in enumerating every difference between open-source Apache and open-source POI.”

Them: “Fine, but what about...”

  1. Bugs?

Me: “Your software is a paragon of virtue?”

  1. Viruses and Malware?

Me: “The other 1,000 users will have found it and fixed it.”

  1. Support?

Me: “You have their source. You have my source. I’m leaving, eventually, you know.”

  1. Licensing?

Me: “Here .”

Varieties of Discomfort . This conversation is really about discomfort with software where the focus is squarely on maintenance, adaptation, integration and support. The bottom line on discomfort is this: “Who do you sue when something goes wrong?”

Everyone denies that “who do we sue?” is the bottom line. But when you provide the license agreement to the CIO and they provide it to the lawyers, the lawyer only looks at a few clauses related to liability. Everything else is – well – technical and not the lawyer’s problem. Once it bounces back to the CIO, it can become a hand-wringing exercise.

Previously, I mentioned [link ] the “I don’t see the value in converting” conversation. This misses the point of open source software entirely, reducing it to a strange either-or proposition. This tend to derail the conversation, leaving the discomfort factor front and center, and ignoring the real issue of cost and benefit.

This plays out into the question “Is it really about replacement or is it about where new development will be done?” [link ]. This question is still way off target, since it provides a strategy for conversion. While a strategic direction (replace vs. evolve) might be a nice thing to have, you can’t establish a direction until after you address the underlying discomfort issue. Clearly, I can’t say this enough times: it isn’t about converting to open source. That implies were converting from something, and reduces the conversation to strange either-or situations.

Buy vs. Build . Tony notes [link ] that the real point is the minute incremental cost in deploying open source software. The incremental cost of another license to use is zero. Support costs do rise with number of desktops served. Indeed, the costs of open source are often comparable to proprietary software, but they’re in-house costs. The point is that we can move beyond the tangential conversations about your level of discomfort by focusing on two things: cost and benefit .

A focus on cost and benefit (and the tiny incremental cost of increasing your usage) leads to the following: What’s the best approach to acquiring the needed software? We used to call this question the “Buy vs. Build” decision. However, it’s expanded into the following bewildering spectrum of choices:

  • Buy and Customize. Possibly the least effective solution: begin with a standardized commercial product, then customize it so that it cannot be supported, maintained or enhanced.
  • Buy . Possibly the lowest risk solution: purchase something standardized from a commercial vendor.
  • Build . In this case, we want something totally unique, and hope we have the technical chops. Emphasis on the hope ; not all organizations are structured to do development well. Indeed, some IT organizations seemed to be intentionally structured to do development poorly.
  • Build by someone else . The strange world of “we want the software, but not the collateral learnings that come from building the software.” Possibly even the worse situation of “we feel the process is more important than most of the products.” [link ]
  • Download . Possibly the most effective solution: download it. Want to customize it? While still a bad idea, it’s more possible here than with something you purchased.

Readiness Test . I think there are two elements of readiness to implement open source solutions.

  1. Desperation.
  2. Focus.

If you are desperate to have your problem solved, you don’t quibble about who you will sue. When I’ve had the frustrating tangential conversations focusing on discomfort, part of the reason is that the customer isn’t truly desperate to solve a business problem. Instead, they’re busy implementing large and complex application software, usually by building it from scratch. They’re proud of their software development capabilities, and bypassing programmers by downloading ready-to-hand solutions isn’t what they want to do.

If you are focused on solving a problem, you can rationally evaluate the fit between a software solution and the problem. If you are doing something else, you have a million other considerations that spin the conversation away from problem -and -solution toward strategy, tactics, operational considerations, who do we sue, and other tangents.

Ramp Up . With all “new” directions, introducing a new set of tools must be done incrementally. It must be focused on solving a problem where the customer is desperate. Adding POI, for example, solved an immediate problem. We had to talk it to death, but the policy against open source wasn’t exactly as broad and silly as the customer initially described. Or, it wasn’t inflexible in the face of a solution and a level of desperation.

You’ll always have a “one true language” conversation [link ], but people will eagerly add Python to the mix when there isn’t a viable alternative. Here’s an example:

  1. We need batch extractions to produce files for a variety of purposes. We have a batch-job production control system with which it must interface. Many kinds of tools can run in this environment, but GUI tools aren’t a good choice.
  2. We need garden-variety CSV and Tab files, and SQL*Plus doesn’t do this without a lot of pain.
  3. We need flexibility, but we don’t want to spend $10K for an elaborate ETL or EAI application. It’s just CSV extracts.

A small technology stack helps speed adoption. Python 2.4 [link ] and cx_Oracle [link ] allows them to write 10-line extraction programs that run from the command-line in AIX or Windows, can be scheduled by their tools, and can be modified without the egregious overhead of a Java application.

One tipping point was the addition of the CSV library module in Python 2.3. I’ve written CSV parsers, and the regular expressions required to handle quote balancing are opaque. In the Python 2.3+ world, we can write a tidy chunk of code that expresses the Database to CSV™ implementation with almost no non-problem technology overheads.

[An example of a non-problem technology overhead is the Java technique of locating a JDBC driver before opening a connection. Yes it’s short, but it’s also confusingly opaque when we only have one vendor for our production databases.]

No Pain No Gain . The cost and benefit of Open Source is only apparent if you have a specific problem to solve. The derailing tangent conversations about “converting” and the “value proposition” seem to happen when discomfort with the approach is larger than the discomfort of the problem itself. If the problem isn’t causing enough pain, then we are free to quibble over solution strategies.

Consequently, my real question is more fundamental. It’s not “why adopt open source?” nor is it “how do we adopt open source?” My question is “Are we ready to support open source when (a) it solves our customer’s problems and (b) they’re ready to adopt?”

Previous topic

The Leading Edge (Revised)

Next topic

The Role of Open Source in the Enterprise IT Investment Strategy

This Page