Before HBase 1.0.0 when a machine in a cluster goes down the data is not lost, If it was replicated using HDFS but can leave critical implications.As before HBase 1.0.0 version a single region was served by a single region server for read and write requests if the RegionServer goes down no read and write requests can be entertained until and unless the automatic fail over recovery completes .
But a distributed system relies on time-outs to discover if something has gone wrong and lowering the heart-beats timeouts to a very small value can result in false predictions while the larger timeouts contributing to unavailability of system until the fail over recovery finishes

The good news HBase 1.0.0 supports the concept of a single region being served by two region servers where One of the replicas for the region will be primary, accepting writes, and other replicas will share the same data files. Read requests can be done against any replica for the region with backup