Today @halter73 gave us a great introduction on the history and future of optimizing Kestrel and how some of the same ideas could apply to data access.
A few topics we touched on:
Usage of managed sockets
Important to reuse handlers (probably already happening with the awaitable we use in Peregrine)
Use new Memory APIs in 2.1: we learned about a few reasons this might perform better that managing the buffers ourselves
Is managed sockets improving further in the long termp?
Usage of pipelines
Simple and efficient programming model for managing buffers with async
There are a few reasons we may want to look at it (e.g. for both Peregrine and Npgsql) even if we feel that the custom code we use right now is ok
Usage of libuv
Currently more efficient than managed sockets for Kestrel, but it is a goal to close the gap
We don't know how much it would impact to use it for data access and this is something we want to experiment with
Usage of new APIs in 2.1
This applies to System.Memory, and the new sockets APIs as well as pipelines
We think it would be ok to cross-compile to take advantage of these if there is significant perf gain
Attendees: @ajcvickers @anpete @halter73, @roji @sebastienros @divega
Today @halter73 gave us a great introduction on the history and future of optimizing Kestrel and how some of the same ideas could apply to data access.
A few topics we touched on: