There is currently an effort on [sFlow.org|http://www.sflow.org/] to standardize a set of metrics for monitoring JVM performance (based on MemoryMXBean, GarbageCollectorMXBean, ClassLoadingMXBean, CompilationMXBean, ThreadMXBean and UnixOperatingSystemMXBean):
[sFlow Java Virtual Machine Structures|http://www.sflow.org/draft_sflow_jvm_2.txt]
An agent implementing this specification will typically run on each virtual machine and periodically send a UDP datagram containing the standard set of metrics to a central collector. Pushing the counters (rather than having to poll for them) is extremely efficient and reduces the operational complexity of performance monitoring in cloud environments where large numbers of JVM instances may be constantly appearing and disappearing.
Currently there is an implementation of the draft specification in the form of a javaagent:
Are there any other standard JMX beans/metrics that should be included? Thanks.