Here is a draft of a Question and Answer page made from material on the
ganglia developers mailing list. There are only a few here, but
hopefully its a decent start.
I brought the webfrontend-install.sgml file in line with the other
install docs. It now tries to hold your hand through the installation
instead of saying "we all know how to install web pages, and if you
don't, maybe knitting is more your thing than computers..." ;) Probably
won't do much good, but the look is more consistent now.
Federico
Questions and Ansers from the Ganglia-Developers Mailing list 6-9/2002.
For use in the 2.5.0 Ganglia release.
Originally by Federico Sacerdoti <[EMAIL PROTECTED]>
Q. How are metrics identified on the multicast wire?
A. Currently ganglia assigns each "well-known" metric its own unique integer
identifier at compile time. This id identifies an incoming metric packet during
operation. Unfortunately, this scheme adversely affects the backwards
compatibility of ganglia, since a new metric can cause the ids of all metrics
to change. When this happens, old ganglia versions will mis-interpret the
"heartbeat" and other special metrics that play an important role in the
ganglia protocol.
In the future, all metrics will be identified by a textual name instead of an
integer.
Q. Why are all metrics in the same namespace? Can there be hierarchical
"groupings" of metrics?
A. Currently, ganglia does not support hierarchical metric namspaces. However,
we have identified several reasons why such a design would be beneficial.
Expect to see fully hierarchical, nested, user-defined metric groups in the XML
output of future releases.
Q. I understand how ganglia adds hosts and metrics. How are they deleted?
A. Although there exists a DMAX attribute for the HOST and METRIC XML tags that
specifies when to delete them, its value is currently always set to zero.
Therefore ganglia will never delete metrics or hosts from its database. For
certain short-lived metrics such as those involved with process monitoring this
shortcoming becomes a significant drawback. To address this issue, Federico
Sacerdoti at SDSC has written a patch to enable and enforce the DMAX attribute.
We hope to integrate his patch in an upcoming ganglia monitor-core release.
Q. Why do we use libdnet?
A. Many things in unix are fairly similar across various OS flavors, whereas
something as simple as wanting to know "what are all the network interfaces on
this box?" or "which network interfaces are multicast-enabled?" is not easy.
Libdnet provides a uniform API to help answer these network-related questions.
We currently use libdnet to query multicast-enabled network interfaces on a
machine, and for metrics such as MTU. However, our tests show the version of
libdnet we use does not support AIX fully.
Rocks Cluster Group, Camp X-Ray, SDSC, San Diego
GPG Fingerprint: 3C5E 47E7 BDF8 C14E ED92 92BB BA86 B2E6 0390 8845