How to configure hadoop to reuse JVM for mappers?

[#1]

Can you explain me how to configure Hadoop to reuse JVM for mappers?

[#2]

Setting the parameter mapred.job.reuse.jvm.num.tasks=-1 causes the

TaskTracker to reuse jvms for the same job. But it's better to start a new JVM for the tasks(if the tasks are not reasonably small) for long running jobs.If we set thse to (-1)it will reused unlimited time.