You may recall I raised this on toasters in early March this year. A quick recap:
We ufsdump some Solaris partitions onto a DLT7000 attached to an F810. They are blocked at 63KB and write performance is not bad (though not quite as good as the dumps/restores of the F810 itself).
But ufsrestore from this drive reads at 10% or less of that rate. Tests showed this problem was not specific to ufsrestore, but seemed always to occur when reading via the rmt(8) interface.
But it didn't use to happen, back when this was Solaris 8 and ONTAP 6.2.2 on an F740, as opposed to Solaris 9 and ONTAP 6.4.3P2 on an F810 - so what changed?
I've now repeated the experiments with ONTAP 6.5.1R1P6, and get the same results.
But, I now have a workround! It turns out that if you stuff more than one "R" (read) command into the rmt(8) connection before waiting for the results of the first one then you can keep the tape moving. Keeping just one extra read pending all the time gives almost the full advantage, and gets ufsdump-type speeds.
So I can run a (Perl) program using that algorithm to pull the whole dump file onto local disc on the Solaris hosts, and then ufsrestore from that. (Luckily, space isn't a significant problem here.) One needs to be careful about subsequent tape positioning because of the extra read(s) past end-of-file that have been done.
I still don't know the answer to "what changed?", though.
Here's an idea that _didn't_ work. I thought: as ONTAP restore reads the tape at a decent rate, why don't I apply it to the Solaris dumps and then move the results from the filer to the Solaris local discs? After all, if Solaris ufsrestore can read ONTAP dumps (modulo occasional problems that have been discussed in other threads over the years), then why shouldn't ONTAP restore be able to read Solaris dumps? So give it a try:
filer> restore rfD nrst0a /vol/test/solaris-test RESTORE: tape file #1 on device nrst0a. RESTORE: Unsupported dump format. RESTORE: Could not initialize media. RESTORE: RESTORE IS ABORTED
Oh well, it was a good idea while it lasted ... :-)
Chris Thompson University of Cambridge Computing Service, Email: cet1@ucs.cam.ac.uk New Museums Site, Cambridge CB2 3QH, Phone: +44 1223 334715 United Kingdom.