July 11, 2014 by Topher . Previously we discussed single-decree Paxos , and then we looked at problems that arise when it’s extended into multi-Paxos.  We concluded with the encouraging suggestion that we could use single-decree Paxos in a different way: the minitransaction. TreodeDB provides atomic writes and yet has no single points of failure, no scaling bottlenecks and no masters to fail-over. Does this sound unbelievable? The minitransaction makes it possible. The Apache Hadoop file system depends on the NameNode to direct clients to the location of file data. A Hadoop cell has only one NameNode, so when it goes offline clients cannot read or write files. The NameNode is a single point of failure . Furthermore, once the number of clients saturates the capacity of the NameNode server, there is no means to divide its work between two or more servers, so the N...