big dd hypothesis

A quick note on methodology:
Tests performed on a Sun Ultra 150, with a Seagate 9 gigabyte SCSI disk.
All times were measured on the "consumer" gtar process.  Buffer cache
was invalidated in between each measurement.  No fastfs was in effect
during any of the measurements.


First we're timing how long it takes to transfer data from one disk to another:
bingy-root> (cd / && gtar --one-file-system -cSf - .) |
(cd /mnt/foo/src && time gtar xfp -)
real    26m2.396s
user    0m19.220s
sys     1m22.860s

A second time, running the same thing:
real    26m9.335s
user    0m17.810s
sys     1m22.810s

...so it's actually somewhat consistent.

Next, we'll time how long it takes to copy disk to disk, without a "big dd":
bingy-root> umount /mnt/foo # invalidate the buffer cache
bingy-root> mount /dev/dsk/c0t1d0s2 /mnt/foo
bingy-root> (cd /mnt/foo/src && gtar --one-file-system -cSf -
.) | (cd /mnt/foo/dst && time gtar xfp -)
real    30m45.378s
user    0m18.080s
sys     1m24.570s

Finally, we time how long it takes to copy disk to disk, -with- a "big
dd", block size of 8 megabytes:
bingy-root> rm -rf /mnt/foo/dst
bingy-root> umount /mnt/foo # invalidate the buffer cache
bingy-root> mount /dev/dsk/c0t1d0s2 /mnt/foo
bingy-root> (cd /mnt/foo/src && gtar --one-file-system -cSf -
.) | dd bs=8192k | (cd /mnt/foo/dst && time gtar xfp -)
real    31m31.881s
user    0m17.100s
sys     1m25.210s

Wow, it's actually a little bit slower :)