Dealing with single point of failure

[Mark]

How does one deal with the fact that HBase has a single point of 
failure.. namely the namenode. What steps can be taken to eliminate 
and/or minimize the impact of a namenode failure? What can a situation 
where reliability is of utmost importance should one choose an 
alternative technology.. ie Cassandra?

Thanks