Running a business on OS X and Linux

In my previous job, I was CTO of a company, MTNI, that sought to put advertising-supported LED signs in transit stations. We had pilot projects with the Cleveland RTA and with Atlanta’s MARTA. We converted both those pilot projects to long-term contracts, but the CEO couldn’t generate sufficient investment to build out the systems, and the company closed down in January 2003, nine months after the CEO came to me and said he wasn’t paying me for the month I had just finished working, let alone the months to come.

Fortunately, the lead investor in the old company had a secured investment, and therefore inherited the company’s assets in February 2003. We sat down, and decided there was a supportable business model, a couple of contracts that were ready to go, and possibilities for future growth.

I learned a lot from the last go-round, and from my time at CNN.com. At CNN, I worked in tech support, which gave me a valuable perspective on a rapidly growing company whose needs were being met by a few “cowboy coders,” who threw together solutions in rapid-deployment systems like FileMaker, Perl, and AppleScript. Then, the “professionals” came in, and started building massive, complicated, Windows-only applications using Visual C++, MFC, and CORBA (success story from 1997; I'm not sure any of the applications mentioned are still in use). The cowboys were sent packing, but the software from the corporatists broke a lot more and was harder to support and troubleshoot. And rapid response? We’ll get that right on the development calendar.

Meanwhile, I built and then managed the development of a fairly complex web-interfaced system to handle support requests and tracking, in FileMaker and Lasso, as a full-time project initially, but later as one small part of my job. The development effort was more cowboy than corporate, so there was one 37-hour day when the product first launched, but it was simple to use, simple to maintain, and simple to debug. It ran on Windows and Mac (later, through the web, on Solaris and Linux), managed daily work for 12 to 30 users, and sent critical updates directly to their pagers. It went down unexpectedly once in four years, when we let its primary drive run out of space (d'oh!).

When I left CNN to the new company, there was a software project already underway, that was to be Java and browser based, but which the largish development shop migrated into a standalone Java app, with Windows support immediately and Mac support eventually. Backending the app were IIS and SQL Server. When they showed us a user interface prototype, and we had changes to suggest, they were almost uniformly “out of scope” (silly me, I thought the client could best define the project’s scope). It wasn't a great surprise when the company filed for bankruptcy before they completed the project.

I'm a Mac bigot from way back. My experience at CNN reinforced my experience supporting the Mac users at Coca-Cola, and in mixed lab environments at the University of Georgia that Macs are (and, even before OS X, were) easier and less expensive to keep running (and most of those studies don’t factor in the most important cost, the cost of lost output for users while some tech is banging on their PC).

MTNI was a mixed environment; we had Linux and Windows servers, Macs and PCs on desktops, and signs running embedded DOS. The Linux servers ran custom Python code that gathered and distributed news, sports, and weather, transit updates, and provided a web interface to the signs. To adequately support the Windows servers, I hired a Windows system administrator I had worked with at CNN, who knew all about Windows, but used it anyway.

The new company is significantly more Unix-y, with a PC laptop for the CFO and the CEO, but Macs for everybody else, and Linux servers running Python and PostgreSQL all around. I gave my developers a choice when I brought them aboard, and they both wanted Macs. They're both new-generation Mac users, with zero or little experience before OS X, and their primary editor and development interface is vi, despite my offers of BBEdit or whatever else. I've bought copies of MS Office for everyone — it's still more work not to have it than to have it.

I hired the same, now former, Windows system administrator, but got him a 15" PowerBook, with Virtual PC, in case he backslides. I don't know if he's even fired VPC up yet, but he keeps dropping things he can't believe are so much simpler on his Mac.

Since they're vi users, I couldn't even see the point in buying the developers G5s, so we saved quite a bit of money by going with 17" G4 iMacs. I spent an hour or two on each of the machines, and I have literally never logged into them since, and none of our Macs has (knock on wood) been down or required any support attention in a combined 3.5 man-years of use.

Since there's literally no work involved to support it, we use Rendezvous to share iTunes, and chat and transfer files over iChat. I wish Address Book had Rendezvous sharing, so we could publish a subset of our contacts. Update 2/4/04: Now you can.Update to Update 2/4/04: Address-o-sync is a neat proof of concept, but we had some issues almost immediately, and I think I prefer the “iTunes” sharing model (pull from whatever computers are visible) to the “iSync” sharing model for contact data.

We've had to spend a good deal of time, and a fair amount of money, building our own interface to manage some signs that also offer a limited Windows interface through a proprietary DLL, but the capabilities we've added by developing it ourself and the deeper knowledge we've gained of our key products more than outweigh those costs.

Python gives us development flexibility, the ability to easily add in other libraries, and fairly simple integration with databases. Since developers tend to write about the same number of lines in whatever language they're using, my programmers have been very productive, and better able to deal with quirks introduced by flaky bosses. For more on the advantages of scripting languages in comparison to traditional system languages, check out John Ousterhout's Scripting: Higher Level Programming for the 21st Century, from the March 1998 edition of IEEE Computer.

I really can't think of anything we've really needed, and been unable to find for the Mac. I've had to convert CAD drawings into other formats to view, but the average business doesn't have PCs with CAD software installed, so I think that's a wash.

Where our 6 Macs have required essentially zero support (again, knocking on wood), the two PCs have not. The CFO had a Toshiba Satellite, and now an IBM ThinkPad, that between them have chewed up several days of support, over setup for her MP3 player, upgrades to her accounting package, and problems upgrading Norton Antivirus. And that's ignoring what a pain it is to switch it onto our wireless network when she visits, between IBM's wireless card interface and XP's wireless control panel, and the need to convert the alphanumeric WEP key we all have memorized into a hex key to log her in. My Mac switches between three different access points, and two wired connections, and I never have to type a password.

I'm not saying that Macs and Linux boxes are no work, but the work, it seems to me, is focused on the solution at hand, rather than the problem of the moment. It looks like I'm continuing to marginalize the Windows boxes — our CEO wants an iMac when we move into our new space next week.

Update: I've posted some more details of how we work.