Hi,
take care do not mix things:
My experience are:
numOfReduceTasks = numOfBoxes
numOfMapTasks = numOfBoxes * n // n = 1...4
Each task processed on a tasktracker run in a separated virtual
machine. So in case you have 4 tasks running on one tasktracker you
to start 4 virtual machines.
I'm no jvm expert but I guess that each jvm has some overhead in cpu
and memory usage. So I _personal_ would only run one per tasktracker
until I use a jvm that does this 'sharing stuff' ... didn't remember
the exact name.
Stefan
Am 21.02.2006 um 17:44 schrieb Byron Miller:
Stefan,
Is this value best set by number of CPU's per
tasktracker node?
--- Stefan Groschupf <[EMAIL PROTECTED]> wrote:
Chris,
first of all it is possible to run several map tasks
in one tasktracker.
See (Hadoop code) TaskTracker.java line 82 also
there is a
configuration value:
<property>
<name>mapred.tasktracker.tasks.maximum</name>
<value>2</value>
<description>The maximum number of tasks that
will be run
simultaneously by a task tracker.
</description>
</property>
---------------------------------------------------------------
company: http://www.media-style.com
forum: http://www.text-mining.org
blog: http://www.find23.net
