The tradeoffs are pretty straightforward. Multiple paths in path.data stripes your shards across the paths (at the shard level), and using RAID 0 stripes your shards across the disks (at the block level). In the first case if you lose a disk, you lose the shards on that disk and nothing else. In the second case, if you lose a disk, you lose all of your shards on that machine. Therefore the first gives you increased safety, but the trade off is lower performance.
It depends on your workload whether or not you need the increased performance, and whether or not you have the risk appetite to accept the reduced safety that comes from choosing the increased performance.
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.