Best practices for choosing fs block size on ext3/4?

I never really pay block size much attention but obviously there can be benefits to choosing something other than the default. I am looking for a good "best practices" paper on choosing block size. Also, when used on top of LVM is it's performance payoffs or importance negated in any way? TIA

From serverfault CarpeNoctem

• Short answer:

  4096

  Longer answer:

  Larger block sizes reduce fragmentation, at the cost of some wasted space at the end of the block. Having said that, give the size of most hard drives these days, unless you are storing something like a mail or news spool you should use 4096 as it matches the page size (in memory) with the block size on disk.

  If you are storing a mail spool, then you should look at using something like XFS which is more efficient at storing large numbers of small files.

  CarpeNoctem : As far as ext3 with LVM do I need to conern myself with the underlying physical volume extents? Do I need to coordinate between them or anything like that?

  Dave Cheney : For a single PV inside a VG probably not, if you add extra PV's to an existing VG, then you'll have to manually rebalance the data across the new volume

  pulegium : +1 for mem page == sector size. really affects performance.
  From Dave Cheney