Reading Brook's last email, I thought I would forward this paper to the
list:
http://www.cs.utexas.edu/users/dahlin/projects/sdims/papers/sdims-
sigcomm.pdf
This work appeared in SIGCOMM 04 (a very good networking conference)
and addresses issues related to our Ganglia system. Their ideas for
handling aggregation are more general and sophisticated than gmetad,
but I am sure also less developed for real-world use.
Their metric registration API is interesting, and more dynamic than
what is currently in gmond. The DHT stuff I'm still working out, I
guess I was tired when reading it for the first time. Anyways, take a
look if you're interested.
Federico
Rocks Cluster Group, San Diego Supercomputer Center, CA
