Wulfstat is a simple ncurses-based xmlsysd client. It provides an instant window onto your cluster from any xterm that can provide the following data views: load averages (most useful all around view); memory (like running "free" on every node); network (rate statistics for all interfaces, per node); times (uptime, clock time, processor type, clock, cache size, and duty cycle); user (key statistics on running userspace tasks, controllable).
wulfstat is configured by means of an xml-based wulfhosts file, which parametrically describes the hosts to be monitored by hostname, ip number, hostname range (so all 256 hosts in b00-bff can be specified by a single tag), and ip range (ditto). This file can also be used to set filters on the user display so only tasks belonging to certain users or with certain names appear.
xmlsysd and wulfstat are still under active development and are not warranted to be bug free. However, they are also being actively used here and bug reports will get rapid attention. xmlsysd itself is fairly stable at this point and quite reliable; the wulfstat client is more rapidly changing and has a few small bugs associated with its development (that will gradually be eliminated) that do not generally interfere with its function but can sometimes cause small delays.
I would welcome contributions or suggestions for xmlsysd and/or wulfstat's future evolution. When wulfstat is stable I've already begun work on gwulfstat, a gtk/glade based GUI version of wulfstat that will likely have similar functionality and features. Given the xml format of the output, it should be a very simple matter to create a web-based client (possibly as simple as defining a suitable DTD and embedding calls in a display update loop). Similarly, it should be very easy to write a perl script to loop over nodes, extract their load averages (or any other measure of interest) from a persistent xmlsyd connection, and execute alarms or send mail according to any criterion desired.