[Gluster-users] glusterfs performance issues

[Jeff Darcy]

On 1/8/13 8:32 AM, Rob Gil wrote:
> Cassandra does this interestingly enough and is more baked than the
> suggestions posed thus far. As it is an eventual consistency database it
> needs to pull the freshest ID of a record, similar to a file. That ID is
> based on a timestamp and a server hash (potentially more things). It might
> be worth looking in to if this is not fully baked.

Actually the use of simple timestamps instead of vector clocks has long been
regarded as a deficiency compared to some of Cassandra's competitors[1].  It's
allowed to persist because Cassandra explicitly does not support that level of
consistency[2], and that fact continues to cause concern among users[3].  Under
POSIX we must support that higher level of consistency, so simple timestamps
are inadequate.

[1]
http://docs.basho.com/riak/1.2.1/references/appendices/comparisons/Riak-Compared-to-Cassandra/

[2] https://issues.apache.org/jira/browse/CASSANDRA-580

[3] http://www.datastax.com/support-forums/topic/vector-clock-conflict-resolution