Stripe unit = 128 * 2048 = 256KB, stripe width = 256 * 2048 = 512KB

If a block write is not aligned on a RAID stripe unit boundary and is not a full stripe unit, the RAID will be forced to do a read/modify/write cycle to write the data. This can have a significant performance impact. By setting the stripe unit size properly, XFS will avoid unaligned accesses.

With an internal log, excessive amounts of log activity can adversely affect the performance of the rest of the filesystem. The external log could also be placed on higher speed storage.

Too few allocation groups limits the ability to parallelise concurrent disk block and/or inode allocations. Too many allocation groups (with little space in each one) will increase the probability of exhausting an allocation group and having to search for another group that has free space. This can cause an unreasonable amount of CPU time to be used when the filesystem is close to full.