
My team uses Hadoop extensively, but while a background with Hadoop never hurts (it's a great system that solves a difficult problem, but it has many quirks and rough edges that need to be worked around) it's much more important that the candidate is a great programmer and has domain specific skills (in data processing, information retrieval, machine learning, statistics, numerical analysis -- the tasks you typically use Hadoop for).
Asking "how do you compute <some function> over <some dataset> on a cluster of unreliable, commodity machines, communicating through message passing" is a much better interview question to use than "what is the difference between TextInputFormat and SequenceFileInputFormat" and "what is the interface you implement to specify a custom partitioner and how do you configure the M/R job to use it".
Matrix Computations by Golub
Collaborative filtering papers by Koren: http://research.yahoo.com/Yehuda_Koren
Stanford paper which is the source of Mahout project: http://www.cs.stanford.edu/people/ang//papers/nips06-mapredu...
Map Reduce algos book by Jimmy Lin: http://www.umiacs.umd.edu/~jimmylin/book.html
Pattern Recognition by Bishop (or Theodoridis)
Linear algebra by G.Strang
Graph Theory by Gross
Hadoop the definitive guide by Tom White
Natural Language Processing by Manning
Information Retrieval by Manning
for additional references here is a great compilation: http://measuringmeasures.com/blog/2010/3/12/learning-about-m...
An example of this is when you search for Hadoop. You'll find all 4 editions of Tom White's book show up in the top 10 listing.
http://www.amazon.com/Hadoop-Definitive-Guide-Tom-White/dp/1...