[impala-user] Impala on existed Hadoop clusters

We are considering how to use Impala in our production environment.  There are two choice:

1.      Building a dedicated clusters

2.      Installing Impalad on the existed hadoop clusters

I prefer #2, but it raises another question, how many Impalad daemons  should I install on existed hadoop clusers? one impalad per datanode?

Yes, one impalad per datanode. We generally recommend to co-locate the Impalads with the HDFS datanodes to allow scheduling queries for local short-circuit reads (better performance).