steven.mccoy at miru.hk
Sat Dec 13 14:09:12 CET 2008
2008/12/13 Martin Sustrik <sustrik at fastmq.com>:
> Yes, it may be interesting to use eventfd internally for waking up 0MQ I/O
> threads. Have you ever tested latency of eventfd vs. pipe/socketpair?
It's on my to do list, as I only have a VM of glibc 2.8. From what I
read on the LKML that it's more of a free convenience from introducing
signalfd and timerfd. A pure pipe vs. eventfd comparison shows a
significant improvement, as the fd count has halved, but real world
applications would barely notice any difference as nothing is
engineered to depend so much on pipe signalling.
More information about the zeromq-dev