appropriate CCD interleave for LFS

[Blair Sadewitz]

Would using a ccd equal to LFS' big block size result in near-optimal
or optimal performance for, let's say, compiling large amounts of
source and/or general desktop use?

The other strategy I've heard of is using [large] prime numbers; is
this more appropriate for ffs due to the fact that it ensures that
inodes are evenly distributed among all components?

Am I mistaken on any point here?


Thanks a lot,

-Blair

[Jason Thorpe]

Post by Blair Sadewitz
Would using a ccd equal to LFS' big block size result in near-optimal
or optimal performance for, let's say, compiling large amounts of
source and/or general desktop use?

For LFS, the ideal interleave (for either ccd or RAIDframe) would
cause an LFS segment to map 1-1 to a row (i.e. to be written to all
component disks in parallel). This is ideal because LFS writes out
entire segments at a time. So, basically, "(segment_size_in_bytes /
number_of_disks) / 512" to get the ccd interleave factor (which is
expressed in blocks, IIRC).

-- thorpej