to show that RTLinux is faster than Linux

Please first define what you mean by "faster":

- Network throughput

- Scheduling overhead

- IPC throughput

- IPC latency

- Filesystem performance

- ...

There are many ways to measure a system's performance, but no general one.

It isn't faster. But it's predictable, you get a guaranteed response within a specified time.

Last time I run such a test (kernel version 2.4.9), I measured interrupt latency by using a fast counter on a custom PCI board. It turned out that the interrupt latency on an unloaded linux system was shorter by a factor of 2 compared to RTLinux on the same setup.

However, as soon as the system load increased, I measured interrupt latencies as bad a 0.5 seconds. The main culprit turned out to be a SCSI driver which sometimes disabled interrupts for that length of time.

In the same situation, the RTLinux interrupt latency stayed unchanged.

Kind regards,

Iwo

Real time is about latency, speed is generally a measure of throughput. A server running a Realtime O/S will typically have lower throughput than a standard kernel, its latency (time to respond) will be faster, but it will not benefit as much from queues and processing large chunks without context switches.

The majority of workloads requiring realtime have only a small realtime element, so its normal to run one process as realtime and the rest of the system scheduled around it - kind of best of both worlds.

You need to analyse the workload, separate hard-realtime, realtime and standard parts of the workload. If you have hard-realtime requirements then you cant "prove" anything - as only a realtime O/S can achieve these.

You need to work much harder on understanding what you are trying to prove before you device a method to prove it .........

Jon