I guess that you are thinking of the VxFS (JFS) block size. VxFS allocates space in extents rather than blocks. The relationship is that an extent must be a binary number of blocks between 1 and 2048 (unless the filesystem is more than 90% full, in which case the maximum extent size is reduced to 128K bytes). Thus for large files, a larger block (or fragment) size allows larger extents and thus more efficient space management. The only constraint on I/O is that VxFS will not allow large I/O operations to span extent boundaries. Small I/O operations are not constrained by the VxFS block size at all.

So if you are building new VxFS file systems for Oracle, then yes, there is a small case to be made for using the largest available block (or fragment) size. However, it is not enough of an issue to warrant rebuilding existing filesystems.

What is the preferred OS blocksize for using filesystems with EMC Storage? These databases are mostly 8k oracle bs, except some datamarts at 16K. The OS blocksize currently is 1k. I had read that they should match, or be a multiple of. My suggestion would be that 8k is better. Other opinions? Any white paper that speaks to this?