hadoop-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Lanati, Matteo" <Matteo.Lan...@lrz.de>
Subject Hadoop network setup
Date Tue, 06 Aug 2013 08:58:23 GMT
Hi all,

the question about the setup of Hadoop with multiple network card has been asked many times,
but I couldn't find the info that I needed. Sorry if this is a duplicate: in this case just
point me to the right documents.

My nodes have two interfaces, eth0 with a public IP and eth1 with a private one. I would like
the users to use the public interface while the private one should be reserved for intra-cluster
communication. At the moment I'm focusing on HDFS.
Having assigned the first interface eth0 to the public network, HDFS daemons listen to it,
so I was able to use the filesystem from any machine. The same interface is also used for
the communication between Namenode and Datanode or among Datanodes involving ports defined
by dfs.datanode.address and fs.default.name. Is it possible to instruct HDFS to use eth1 for
I tried to play with the following options:

- dfs.datanode.dns.interface: it only affects the DFS node status page published by the Namenode
- dfs.client.local.interfaces: I tried to set this up on the Datanode, but I didn't see any

Finally, can you please explain the role of the hdfs-site.xml option 'dfs.datanode.ipc.address'?
What is this IPC server used for? By default it is set to 50020. I tried to dump the network
traffic while performing some operations but the port was never used.



Matteo Lanati
Distributed Resources Group
Leibniz-Rechenzentrum (LRZ)
Boltzmannstrasse 1
85748	Garching b. M√ľnchen	(Germany)
Phone: +49 89 35831 8724

View raw message