A couple questions about monitoring server performance:
If your server is so overloaded that you can't even get input to the shell, you can't run 'top'...or sometimes you have 'top' running, but it barely updates, if at all, because the server is so overloaded. Do you have any way of determining what is slow and why, other than restarting the server (or waiting it out if possible?) As in, is there any other way to get an idea of what the performance characteristics of the machine externally?
Along the same lines, do monitoring programs like Cacti have any way to determine the performance characteristics externally (and thus stay accurate even when the machine is completely overloaded)? Or does the accuracy of monitoring programs like this break down under extreme load?