Thanks Devaraj Setting multiple entries to fs.data.dir conf makes multiple datanode threads, and works fine. However they are SubThreads, so only one connection to namenode. It's some different from real cluster. Now I start a datanode with one dfs.data.dir conf value and then start a second datanode with another dfs.data.dir conf value, it works fine too and there are 2 connections to the namenode now.
On 7/25/06, Devaraj Das <[EMAIL PROTECTED]> wrote:
Moved this to hadoop-user... My 2 cents: A datanode is identified by its storageID. If you have multiple storage devices in your box (visible to the user as different directories) then you can have multiple datanodes on your box (one for each directory). The way to do that is to specify a set of comma separated entries in your conf file for the dfs.data.dir conf value. These datanodes will bind to different ports. -----Original Message----- From: yichao.zhang [mailto:[EMAIL PROTECTED] Sent: Tuesday, July 25, 2006 2:12 PM To: [email protected] Subject: Is there a way to start namenode, 2 datanodes on one box? Hi is there a way to start the name server and more than 1 data servers on just one box. I have only one PC and want to test against dfs. I have set up one name server and just one data server in psuedo-distributed mode on my box in eclipse. However it seems no way to add another data server even if I have 3 Ethernet adapters, one is a real d-link card, the other 2 are Microsoft loop adapters. so my box has three different ip. but the data server is identified by machine name, so there is impossible to have 2 data servers on a single box, is that right? or is it possible to simply identify a data node by its ip to solve this issue? your 2 cents appreciated! -Yichao
