The nice thing about running it on mesos is you can share the same machines with other big data apps like Hadoop or spark and don't have to provision separate machines.