[rabbitmq-discuss] RabbitMQ C library function amqp_simple_wait_frame takes 400 ms

[Tony Garnock-Jones]

Hi Jim,

I'm glad TCP_NODELAY worked for you in the end! :-)

Jim Irrer wrote:
> I suggest modifying the C client to set TCP_NODELAY in the
> amqp_open_socket function.

I think that sounds reasonable. After all, what we're doing is closer to packet
transmission than stream transmission, by and large, so choosing packet-y
defaults is probably more reasonable than TCP's stream-y defaults.

[...]

I've just tried it out, and it does (as expected) reduce straight-line
throughput a little, but I think as you say, people who need straight-line
stream-like behaviour can always switch TCP_NODELAY back off again.

I'll commit this on a branch -- bug22931 -- and we'll see what the rest of the
team thinks of it.

Thanks,
  Tony