Highway POS System
Why Linux? Because it works. Because it does the job. Because, let's face it, it's free.
In late 1993, we started doing preliminary design work on a new retail petroleum point of sale (POS) unit. Basically, it's an electronic cash register (ECR) which also handles credit cards and controls fuel dispenser systems. We wanted to design a PC-based system to leverage the ease of availability and low prices associated with PC hardware.
Our sister division in the Netherlands had already began designing a new POS system for the European market. Since they had provided major components of our current successful products, we waited in heady anticipation for their design documents. Unfortunately, their design was based on MU-DOS, a multitasking DOS by Digital Research. We weren't very happy, and started researching alternative operating systems.
We had the following requirements:
Soft real time: While we didn't need a true real time system, we did need an operating system which would be able to handle over a dozen serial ports. These ports weren't always active, but we'd have bursts of communications from 9600 baud and higher, and we needed to be responsive in processing the incoming data. We also had to handle an experienced ECR cashier who would be entering keystrokes at a high rate and expecting a quick system response.
Multitasking: Preferably preemptive. What we really wanted was protected memory segments, so that one task wouldn't be able to corrupt another's data space. Being able to use multiple programs would allow us to change out specific programs without affecting the rest of the system.
Non-proprietary: We wanted an OS and development system that was supported by an outside agency using industry standard tools and utilities. We wanted to make our own workload as small as possible.
Inexpensive: The price point we were aiming at for the product didn't allow too much choice in the way of software.
In fairly short order, we had examined a number of candidates and decided to use SCO Unix w/Chorus Micro-kernel. Linux hadn't become an option—yet. We had heard of it, but didn't feel it was mature enough or commercially acceptable enough to warrant serious investigation.
Windows 3.1 still seemed too buggy, Windows NT was too expensive and too much of a resource hog.
QNX suffered from a lack of virtual memory. While we didn't have a real need for virtual memory, we wanted the ability to use it if the application needed it for some short but highly memory intensive operation.
OS/2 was originally high on our list, as our sister division was also looking at migrating to OS/2 for their product. Unfortunately, we consistently had problems with installation and found support extremely difficult to get.
SCO's only real problem seemed to be pricing. They wanted much too much, much too soon. While we started negotiating, we needed to start developing. We didn't want to purchase any SCO development systems, so we decided to start our development on Linux. Linux was fairly POSIX compliant, so we were confident we would be able to port our code without difficulty to another platform, particularly another Unix platform.
Differences in usability showed up almost immediately. I was working on the one SCO system we did have (an evaluation unit) and found the utility commands to be almost primitive compared to Linux. Things as simple as the recursive (-R) flag were missing from the chmod and chown commands. I found myself locating and acquiring more and more GNU tools. I was unable to get good or timely support for the SCO C compiler and the C++ front end. (They didn't have a native C++ compiler at the time.) I had difficulty in getting quality technical support. I had plenty of their time, but kept talking to support personnel who were not close enough to the module for which I needed information. I continually had to bother my SCO sales representative to renew my support contract (which was free due to the on-going negotiations).
On the Linux side of the partition, life was much smoother. The utilities were very advanced, the documentation reasonably good and the Linux newsgroups were a wealth of information and solutions. We tried using the SCO newsgroups, but found that, unless someone had experienced the exact same problem, little help was forthcoming. Linux newsgroups allowed us to contact the actual authors and maintainers of specific areas of code to better understand and troubleshoot problems. They were also fast. Within a matter of a day or two (sometimes less) we had fixes, workarounds and suggestions to try. It was almost like having a bunch of Linux experts on our payroll.
While upper management continued negotiating with SCO, engineering started quietly hoping that we could stay with Linux. However, we still had that commercial issue. We were afraid our customers wouldn't accept a free “hackers” OS. Luckily, Linux just kept gaining momentum. Linux 1.0 came out during this time, and suddenly companies like Caldera were making this wonderful OS a commercial product.
Suddenly, using Linux wasn't quite the issue it had been the previous year. Our customers still weren't sure, but their technical people started giving grudging acknowledgement that Linux just might be okay in a product. We decided to stay with Linux and drop SCO, who still wanted too much, too soon—we've never looked back.
Currently, we've got Linux 2.0, demand loadable kernel modules and connectivity to Windows, Novell and AppleTalk. Most importantly, we never hear the words, “Hmmm, it looks like an OS problem.” We still have the occasional problem with an application or utility, but it's rarely anything critical, and, if it is, we always have the source code and access to the author.
We aren't the only retail petroleum POS vendor to come up with a PC-based solution. We are, however, the only Unix-based one. There have been attempts based on Windows 95, OS/2 and even DOS. So far, no one's really succeeded. Our product will hit the streets around mid-year (1997). It's already done wildly better than expected in its first field trials. We have confidence that it's going to be a winner, due in large part to Linux.