On Wed, Mar 2, 2011 at 2:51 PM, Peter Schuller <peter.schul...@infidyne.com> wrote: > Cassandra is about as friendly as it can be with respect to > fragmentation. All data is written sequentially and in bulk.
That said, there can be more than one thread writing at a time, which doesn't help fragmentation. On the plus side, "more than one" is never more than a handful, which is relatively fragmentation-friendly. Potential sources of simultaneous disk write : - Memtable flushing - SSTable compaction/anticompaction - SSTable streaming - Commitlog writes (which should be on a different drive from data per best practices anyway..) =Rob
