Jens Axboe

I'm just not a fan, most of the commentary (to me) are from folks looking to look smart and not knowing a lot of the details. In many ways, not...

Tried to compile this as I'm pretty convinced something is amiss with the single thread performance, but it fails for me: ``` Consolidate compiler generated dependencies of target photon_obj [...

OK got it going, and the examples built. signal.ccp is missing a pthread.h include.

> juxtaposing the performance variance between epoll and io_uring for 512 + 1 client in test 1... vs equivalent performance in test 2 with that usleep... my intuition is all...

Got it built and running, but there are no docs on how to run with the various backends on either the client or server side. The interval thing doesn't seem...

With the actual instructions, I gave it a test spin. From a quick look, you're doing a lot more on the io_uring side than you are on the epoll side....

Another note - lots of receives will have cflags == 0x04 == IORING_CQE_F_SOCK_NONEMPTY, meaning that the socket still had more data after this receive? Is this really a ping-pong test,...

We're also spending a ton of time in __vdso_gettimeofday() when run with io_uring, and I see nothing if using epoll. This is about ~10% of the time spent! It's coming...

I don't understand this issue. You're using an unknown option, and fio helpfully tries to tell you that you probably meant ```--enghelp```. ??

Since you already root caused the problem, why not just send a patch or PR to fix it?