[Beowulf] IEEE 1588 (PTP) - a better cluster clock?

Ashley Pittman apittman at concurrent-thinking.com
Tue Jul 24 07:15:14 PDT 2007


On Fri, 2007-07-20 at 16:41 -0700, Jim Lux wrote:
> 
> There are tons of reasons why one might want more precision than NTP 
> can provide, leaving aside the obvious applications for things like 
> controlling test equipment. 

But none of the ones you list are more than vaguely related to HPC.

Personally I'd love it if clocks in clusters were better than they often
are now but that's because a surprising number of people seem to think
they can live without ntp and are content with living with large
discrepancies across the cluster.

The only thing I've found ntp to be lacking in is the ability to compile
complex software over NFS on 2.6 kernels which have 64 bit timestamps,
mpich springs to mind as the configure step often enters an infinite
loop.  This isn't helped by some of the cheaper NAS offerings not
supporting ntp at all.

In short, I'd use it if it was packaged with the distro and just worked
out the box (without using multicast preferably) but it's absence isn't
enough to cause me problems.

Ashley.




More information about the Beowulf mailing list