A DESCRIPTION OF THE REQUEST :
Creating jmap dumps blocks JVM for considerable time. During this time clusterware like jgroups or caches might experience problems.

This might be omitted in case jmap would first issue a "block" then would fork the entire process. The main process could run like nothing happened, the forked process could GC (if needed) and dump the data.

On a fairly busy machine this would mean that the system needs memory to hold two processes, but this is ok most of the times and I would happily accept it for safer jmap dumps

JUSTIFICATION :
JMap from production systems with huge heap is not feasible currently. Figuring out hard-to-find problems is limited.
With forking this would be much risk free.