Conversation
Notices
-
I've been trying to copy a single 4TiB ZFS dataset between 2 machines for > 1week now.
On dedicated 1Gb/s network with jumbo frames it takes ~ 20 hours over SSH ... however SSH keeps dying after ~ 12 hours with "Broken Pipe" :-(-
`zfs send tank/foo@then | ssh user@backup zfs recv tank/foo` fails
`zfs send tank/foo@then > foo.zfs; scp zoo.zfs user@backup:/tank/` fails-
Anyway, I had enough space on the local box to dump the stream to file; and have copied it remote using netcat
`zfs send tank/foo@then > foo.zfs`
`cat foo.zfs | nc backup 9000`-
Now to see if the stream will import at the other end.
-
I guess 4TiB will not seem like very much data to some people. But it's like 3,000,000 floppy disks, and it's a single file. - remember how much of a chore copying floppy disks was ? In 20 years, it'll be like "OMFG remember when you thought copying 4TiB was hard, lol"
-
https://indy.im/attachment/1110 4TiB data copying
-
-
It's imported! woot...
now can send incremental snapshots; just the daily diffs. Life gets easier!
-
-
@boneidol I'd do "md3sum foo.zfs" on both sides, just to make sure. Sometimes systems with long uptimes get a bit weird.
-
@mcscx what is md3sum ? I can't find it in FreeBSD ? I think I will do a md5 of the files though. They sizes are the same, it's just everything takes *so* *long*!
-
@boneidol s/md3sum/md5sum/ :-) I'd carry on with the import but keep the 2 file and do the md5sum over night. If possible.
-
-
@mcscx everything in ZFS is checksummed. https://blogs.oracle.com/bonwick/zfs-end-to-end-data-integrity but if it fails to import, at least I might be able to work out where the failure happened.
-
-
-
-
-
-
-
