On 1/11/10 9:23 AM, psdc1978 wrote: > Hi, > > I've some questions about hadoop MapRed architecture: > > 1 - It only exists one TaskTracker to one JobTracker?
Pedro: There is a one JobTracker to many TaskTracker relationship. Generally, all slave (worker machines) in a cluster run task trackers. The job tracker is what clients will submit jobs to. I highly recommend the Cloudera intro to HDFS and Mapred videos[2] or the Hadoop wiki[1] and the getting started guide for the overview of the architecture. > 2 - The Tasktracker and the JobTracker are two different instances that > are started only through the start-mapred.sh script? The start-mapred.sh script will start a job tracker on the node you run the script on, if I'm not mistaken. It also ssh's to each node listed in conf/slaves and starts a task tracker there as well. > [snippet of start-mapred.sh] > "$bin"/hadoop-daemon.sh --config $HADOOP_CONF_DIR start jobtracker > "$bin"/hadoop-daemons.sh --config $HADOOP_CONF_DIR start tasktracker [1] - http://wiki.apache.org/hadoop/ [2] - http://www.cloudera.com/hadoop-training-mapreduce-hdfs Hope this helps. -- Eric Sammer [email protected] http://esammer.blogspot.com
