When working with a large corporate site in 1997, an engineer built a test bed to validate the capacity for a Linux-based solution to handle a production load in a 30-user CAD shop. The server was also used to host a project tracking application and accounting application. He built two identical machines, one running Linux and the other running Unix. He compared memory usage, the user and system load factors and responsiveness on both machines as the load was increased on each system in turn. Both machines had dual PentiumPro CPUs.
He examined the number of concurrent client tasks that could be run before the load would become excessive. The benchmarks for excessive delay were set at:
- CAD file save using Samba: 60 seconds
- Accounting application module recompilation: 5 minutes
- Search through entire project database: 2 minutes
Unix Linux ---- ----- CAD file saves 5 8 Recompilation 3 2 Search database 7 11When the results were reported, the head of IT determined that these results confirmed that Linux was not sufficiently scalable. The engineer who worked on this had concluded that Linux had out-performed Unix in the two areas that mattered most. Precisely what scalability meant made could not be established.
In a second case, a site installed a Web server (late 1998) that was supposed to be capable of taking the load of 2 million hits per day. The software vendor had recommended a Sun server. The site engineer installed a dual processor Pentium-II system with lots of memory. The machine comfortably handled the real load, never using SWAP and net ever exceeding a load factor of 3. When the head of IT resigned, the first action taken by the new head of IT was to replace the Linux operating system with MS Windows 2000 Advanced Server (late 1999), so-called because it was more enterprise ready and more scalable.
So precisely what did scalability mean in these situations?
Clearly these examples are biased! But they demonstrate the point that emotional and non-technical factors often prejudice perceptions of scalability.
This was first published in September 2003