>At the risk of boring other readers, I'm curious what numbers you were
>getting during your test - I just tried a --stats myself on a 51MB

rsync version = 2.4.4

Working with a 100MB file and doing a null sync, I see:
rsync --stats -e ssh -a --block-size=64000
wrote 9866 bytes  read 6659 bytes  158.13 bytes/sec

rsync --stats -e ssh -az --block-size=64000
wrote 9866 bytes  read 103 bytes  73.57 bytes/sec

Extrapolating to 20GB, I get almost exactly 2MB.  It seems I slipped
an order of magnitude earlier which makes the need for implementing
the irsync algorithm a little weaker.  

I would still like to see irsync implemented because it would reduce
network utilization significantly at the cost of maybe a bit more cpu
when there are few changes.  It would also help when there are many
scattered changes because it avoids the fragmentation problems that
arise from having a constant blocksize.


Reply via email to