kdb-ts

Description

Event timer library for kdb. Provides multiple high-efficiency implementations of timer loops, which are picked based on availability and suitability to operating system.

• Linux: timerfd. Events implmented as sd1/read() on timerfd. timerfd allows for nanosecond resolution of event specification.

• Default handler: Uses q's system["t"] api to specify the next event timer. Allows for millisecond resolution.

Notes on resolution vs accuracy

There are several issues to address here.

True nanosecond resolution timing sources remain the exception rather than the rule and the usual interfaces used in modern operating systems stop at microsecond resolution at best. So any applications requiring very high resolution should bear this in mind.

Accuracy at these timescales is also a very hard problem: using the best-in-class sources (e.g. GPS hardware) along with well implemented NTP or PTP systems, you may be able to maintain clock sychronization to <1ms, perhaps even a few hundred microseconds. At this timescale any concept of 'absolute time' between sources becomes a bit hairy and you should account for this in your design also.

Finally, whilst I am confident in the design of the API and event loops within this code, it has not been designed with hard real-time constraints in mind. In particular:

• There has been no attempt to ensure fairness or to avoid priority inversion between events; patches welcome if you happen to know how to do this.

• Event lookups will display linear complexity (at best!) with number of events, regardless of frequency of calls. It's possible that this could be improved by using a hash lookup on event ids and looking up the next event id, reducing most maintenance logic to constant-time lookups and updates. I did consider making the events table a sorted step-lookup but did not have an obvious solution to the problem of having two events with identical timers. This also leads me to the next consideration...

• I have developed this library from the outset to prefer simplicity over cleverness. Right now this is a pet project for me and I want to enjoy it :-) Changes will be made based on their maintainability as much as possible.