| |||||||||||||
TALPA strides forwardTALPA strides forwardPosted Aug 28, 2008 11:05 UTC (Thu) by evgeny (guest, #774)In reply to: TALPA strides forward by nix Parent article: TALPA strides forward
I'm not sure. Consider distributed compilation farms (here "distributed" may refer to either filesystem and/or compiler; or just an NFS-mounted volume in the simplest case). Then maintaining nanosecond-accuracy time sync between several computers is needed, which is not trivial.
(Log in to post comments)
TALPA strides forward Posted Aug 28, 2008 11:58 UTC (Thu) by nix (subscriber, #2304) [Link]
NTP can already report jitter and offset values. Maybe what we need is a
way to have those values *reduce* the precision of the kernel-provided nsec timestamps, so that you get timestamps as accurate as possible for your timebase, but no more accurate? (Of course, if the jitter changes a lot, interesting things may happen, but that's quite rare.)
TALPA strides forward Posted Aug 28, 2008 16:01 UTC (Thu) by bfields (subscriber, #19510) [Link] Note that no linux filesystem has time resolution better than a jiffy. (The on-disc format may use nanoseconds, but the mtime/ctime/atime aren't updated using a nanosecond-precision time source.)
TALPA strides forward Posted Aug 28, 2008 18:43 UTC (Thu) by SEJeff (subscriber, #51588) [Link]
"""Then maintaining nanosecond-accuracy time sync between several computers is needed, which is not trivial."""
It is actually really easy if you are not using Cisco switches. The latency of the switches makes a big difference.
Use ptpd from the linux hosts:
It will allow you to keep nanosecond time sync of all machines in a lan using multicast.
|
|||||||||||||