Ferenc Bodon, Head of Benchmarking at KX, recently posted an interesting LinkedIn blog regarding research he conducted on simultaneous execution of functions across multiple kdb+ processes.
The work stemmed from an analysis of the speed and scalability profile of coordinating multiple workers from a central controller in storage efficiency tests using the KX Nano performance tool.
The analysis looked at two different methods for coordinating the worker processes: one using inter-process communications and one using file operations. For the former, he looked at a number of different approaches using each and peach functions for managing process coordination over multiple connections. Variations included using async flush to address blocking, broadcasts to reduce message serialization cost and a timer for coordinating simultaneous starts. For the file-based approach, Linux’s inotify function was used.
Analysis of results over multiple runs shows that timer-based approach delivers the best and most consistent results. It should be noted, however, that the optimal trigger offset depends on the hardware and network environment it’s running in.
The peach handlers approach, which manages the synchronous connections automatically, offered comparable performance but carries the overhead of dual operation (and maintenance) on both the controller and the worker processes. The peach one-shot approach, which manages available connections asynchronously, offers an expedient balance between the two, delivering speed and resilience without the overhead of complex configuration, implementation and maintenance.