For a relatively small, one terabyte cluster ( 2TB actual after replication ) I was trying to nail down what the namenode's ideal memory/cpu size would be, having worked with hadoop off and on as an end-user I can't imagine it being too crazy... but googling for hadoop minimum requirements hasn't given any definite answers or even a recommend server setup.
It's going to depend more upon how many files you have in HDFS. The rule of thumb is that the namenode process itself needs about 1GB of memory per one million objects (files, directories, blocks) held in the DFS. The CPU needs should be marginal.
https://twiki.grid.iu.edu/bin/view/Storage/HadoopUnderstanding has a decent overview of some of these things to take into account.