On Thu, Aug 31, 2006 at 07:25:08PM -0600, Ian Cunningham wrote: > I have filed bug 114 > <http://bugzilla.ganglia.info/cgi-bin/bugzilla/show_bug.cgi?id=114>. I > have also provided a patch. > > The jist of the bug is that we have these new machines, with new > versions of redhat, that scale the processor speeds like laptops do, in > order to preserve power. gmond reports the speeds of the cpu's on these > machines way too low because they are idle when gmond is started. > > The patch is to display the maximum frequency for one of the machine's > processor, instead of using what is (mis)reported in /proc/cpuinfo. With > the frequency scaling feature enabled on multi cpu machines, it is > possible for one processor to have a different speed than the others in > the machine. The question I have for developers is what should gmond > display about the cpu speeds? Should it display: > 1. the maximum speed for all processors [1 number] > 2. the current speed for all processors [1 meaningless number] > 3. both the maximum and current speed for all processors [2 numbers] > 4. the maximum speed for each processor [n x 1 numbers(s)] > 5. the current speed for each processor [n x 1 number(s)] > 6. the current and maximum speed for each processor [n x 2 numbers] > 7. other (display the frequency range) > > My vote is for #1, because its the easiest to code and simplest to > understand. Sometime in the future we may have machines that have > multiple processors with different max speeds. I'm not really asking for > your vote though, more for your opinions. I don't really see this as a > big issue, but one that should be considered.
Given the current limitations of the Ganglia schema, 1 is the only rational option. That's what I've implemented in FreeBSD (to the extent I can since the data isn't always available). Eventually we may want a more complete model of CPUs where we can specify the current and maximum speed of each core along with information about packages, core, and threads of execution per core but that's beyond the current Ganglia design and I haven't seen anything I really like yet. -- Brooks
pgphFhnOm4ZBj.pgp
Description: PGP signature
