I am the maintainer of some legacy code in my company (sigh). The code uses a fair bit of multi-cast even though 98% of applications are all running on same box. It was done this way to distribute applications across boxes should the need arise to scale the system.
These components are talking to each other by way of message passing (some applications have a publisher, other subscriber waiting for the messages to arrive).
My problem - I have some regression tests where some messages are expected to arrive to an application from some component in next N seconds but they don't at times.
To me, this whole approach looks wrong. If the system is loaded then this makes no sense. But I have to fix this and what I am looking at is kernel tuning - better system response by fine tuning the kernel. Is this the right approach?
