South African Business Uses Linux to Connect

The story of a company replacing Windows systems with Linux to obtain better speed and greater reliability.

Linux is the best tool I have found for facilitating the expansion of Internet and Intranet systems throughout the company, Stocks & Stocks I.T., that employs me.

Upon arriving in South Africa from Australia, my first task was to develop and implement a system which would allow general Internet users to obtain accurate, real-time hotel room availability details from the company's chain of hotels and game resorts. This is a list of what I had to work with:

  • WAN spanning South Africa

  • System protected by firewall

  • TCP/IP as the primary protocol

  • Motorola routers

The immediate problems to deal with were:

  • Each hotel/resort used Novell Netware 3.12 running only IPX/SPX.

  • Each hotel/resort room database was approximately 300KB in size.

  • Each hotel/resort was linked using a 64KB line (at best).

  • Everyone was running around with Windows in their mind.

Getting Started

The first prototype system I set up was a central Linux web server as follows:

  • Pentium 133 64MB RAM, S3-Trio 64 with 4MB RAM, ADI 17-inch monitor (Microscan), 1.2GB EIDE HDD, 8x CD-ROM, 3Com 3C509b NIC (network interface card), Ergo keyboard/mouse (Microsoft)

  • Linux Slackware 1.2.13, NSCA HTTPd v1.10

The reason for having 64MB RAM, a 17-inch monitor and other “niceties” was so this system could also be used for web development of any home pages/Intranet pages.

At each of the hotels the only system available to me was Windows 95 on which I used a scheduling system to send the room database to the central web server every morning at about 6AM. The Windows 95 machines were also being used to act as a bridge between Novell and TCP/IP. This gave the Internet user statistics on room availability that could be up to 24 hours old. The room databases were in DBase-IV format, for which I wrote a C application to search and return statistics given desired room and date details.

The system seemed fine for a while, until the Windows 95 machines mysteriously started to shut down, lock up, etc. for no apparent reason. It was also about this time that the “higher management” decided that the ability to obtain “real-time” room availability statistics was a priority.

I felt real-time statistics would be an impossibility with the current setup, as it meant sending 300KB of data to the central Linux server over a 64KB line in less time than it takes for the average person to get bored and move on to another web page.

Current Prototype

For the current prototype, the whole process was turned upside down. While still retaining the central web server, I replaced the Windows 95 machines with small, highly effective Linux machines (2.0.29). These Linux machines now process availability requests locally via a secure shell from the central web server.

In order to provide a backup, I left the existing Windows 95 system on each machine and used FIPS (First Interactive Partitioning System) to create a new partition on which I installed Linux. Details of the “remote” machines are as follows:

  • Pentium 100 (Generic Clone), 8MB RAM, 800MB HDD, 3Com 3C509b NIC (no monitor, no keyboard, no mouse)

  • Linux 2.0.29 (Slackware), Secure Shell (ssh), ncpfs (network file system that supports the NCP protocol used by Novell)

ncpfs is used to mount a Novell Netware volume containing the room availability database to a local directory at boot time. It should be stressed that without ncpfs, there is no way this project would have gotten into the “real-time” phase successfully.

The process of checking room availability for hotels is now as follows (Local = central web server, Remote = Linux system at required hotel):

  • Local: Get room/date details (HTML POST form).

  • Local: CGI application processes POST data and executes.

  • Local: Use ssh to log in to the appropriate remote Linux machine.

  • Remote: Search database located on Novell Netware server.

  • Remote: Exit search with 0 = Success, 1 = Failure.

  • Local: ssh returns the result of the search through its exit value.

  • Local: CGI application generates an HTML file with the correct response, depending on the database search result.

This whole process has been timed to take less than 10 seconds which I believe is an acceptable figure (given that there is a great deal of existing traffic on the lines). Something worth noting is that using ssh in order to remotely execute processes greatly improved operations. ssh offers public key encryption for all transactions and is effectively easier to setup than rsh.