Hi, For anyone who is interested in installing Yahoo patched hadoop 0.20.2+ ( branch-0.20-security) in rpm or debian form. There is a new option. There are proposals from Owen to standardize Hadoop deployment in https://issues.apache.org/jira/browse/HADOOP-6255. I created rpm/debian packages from the patch, and it's two rpm/dpkg install commands to get a single node cluster running on RHEL/CentOS 5.5/Ubuntu 10.10.
rpm -i hadoop-0.20.100-1.i386.rpm rpm -i hadoop-conf-pseudo-0.20.100-1.i386.rpm If you want to deploy a real cluster, edit configuration in /etc/hadoop without installing conf-pseudo package. Pig and Zookeeper also have packages prebuilt for them. HBase and Chukwa packages are also under way. The test packages are located at: http://people.apache.org/~eyang/ Hopefully, the patch will be part of Apache, and we can download packages directly from Apache in the next release. Feedbacks are welcome. Regards, Eric On 2/23/11 5:01 AM, "real great.." <greatness.hardn...@gmail.com> wrote: Thanks a lot.. On Wed, Feb 23, 2011 at 3:12 PM, Eric <eric.x...@gmail.com> wrote: Cloudera offers a nice distribution and decent documentation of how to install. When you get started, start using the "old", deprecated API as others have pointed out before. It's most complete and most stable for now. 2011/2/23 MONTMORY Alain <alain.montm...@thalesgroup.com> Hi, For my point of view it is not a trivial question... The latest "stable release" is 0.20.2 (embedded in cloudera CH3) (and not 0.21)... When you start with hadoop recently (end 2010 for me) you are facing "old API" depreceated, so you start with using new API... But in 0.20.2 not all the new API are available under mapreduce (example MultipleInput is not available), so you try 0.21 version where it is available... But the 0.21 seems for me not very stable (we are facing a "null pointer exeception" in framework logs without any idea to solve it), so we scope down to 0.20.2 and we are using "Old API". search " Re: Which version to choose" in the mailing list and follow the advice of Todd Lipton. The "old API" are not so depreceated, they will be supported for years because there is thousand jobs running on them. The "new API" could be used when a stable release will be up (0.22, 0.23..). It is the feeback of my personal experience where i lost time trying to use the latest 0.21 version... Since i use cloudera 0.20.2+320 with old API and i don't have any problem (we are also using Cascading to simplify MR writting with very little overhead on performance (6%) versus native hadoop MR jobs. Overall we gain 4,65 factor versus traditionnal RDBMS approach.... Hopes this help you, regards Alain [@@THALES GROUP RESTRICTED@@] De : real great.. [mailto:greatness.hardn...@gmail.com] Envoyé : mercredi 23 février 2011 04:42 À : mapreduce-user@hadoop.apache.org Objet : easiest way to install hadoop Hi, Very trivial question. Which is the easiest way to install hadoop? i mean which distribution should i go for?? apache or cloudera? n which is the easiest os for hadoop?
