Closed threonorm closed 6 years ago
I believe interrupts are not working on AWS right now. That is the reason why we currently need this active polling. That will sacrifice one thread to that active polling on AWS, but as an intermediate solution waiting for interrupts, that makes the speed much more acceptable.
Deleted 1 character, sped up 10x :) Need to fix AWS interrupts
On Thu, Nov 9, 2017 at 7:57 PM jankcorn notifications@github.com wrote:
Merged #146 https://github.com/cambridgehackers/connectal/pull/146.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/cambridgehackers/connectal/pull/146#event-1335214932, or mute the thread https://github.com/notifications/unsubscribe-auth/ACU3s6gQMgqysqNav7Ty1j3-utlNyn3Wks5s05-WgaJpZM4QYe9U .
Yes, it is a ridiculous timeout and will change it back once interrupts are debugged.
I am not sure why you want to reduce the poll timeout interval down to 1ms. This will cause the 'event indication'/listening thread in the host program to pop back back and forth between user space & kernel when an indication has not been raised. (If an indication is detected, then the polling will immediately return to the user, without waiting).
I don't object to accepting this change, but am just wondering if the core problem might be dropped PCIe interrupt indications. If these indications are being dropped on AWS, then the change is perfectly fine (otherwise, interrupt latency is quite long).
Thanks for looking into this!