[zeromq-dev] 0MQ Tests on Linux Real-Time Kernel
Steven McCoy
steven.mccoy at miru.hk
Thu Oct 23 16:18:38 CEST 2008
I started development on the RT kernel as it was the only platform for high
resolution timers, since then the mainstream kernel has introduced
sufficient support for it not to be necessary for regular usage.
High resolution timers are necessary to implement a general low-latency
recovery mechanism, low as when compared with Rendezvous, pre-version 8,
recovery times which rely on 4ms resolution timers. Using TSC has proven to
be problematic on multiple AMD core and HT-enabled Intel systems due to
clock drift.
This should mean OpenPGM has a higher reliability than Rendezvous 7, but
there are no tools to sufficiently quantify messaging reliability in
numerical form. It would be nice to have a signal-to-noise-to-fanout ratio
for IP Ethernet, where you could clearly see multicast and router-assist
taking an advantage at larger scales.
--
Steven McCoy
2008/10/23 Martin Sustrik <sustrik at fastmq.com>
> Btw, have you ever tested OpenPGM on RT kernel?
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20081023/0b0c5e29/attachment.htm>
More information about the zeromq-dev
mailing list