[Beowulf] OT: LTO Ultrium (3) throughput?

David Mathog mathog at caltech.edu
Fri Jun 27 13:32:32 PDT 2008

Short story:  There was an HD68 SE terminator on the tape drive.

Long story:

After messing about with block sizes the highest throughput seen was
38.7 MB/s.  Possibly a coincidence, but that seemed awfully close to
40MB/s such as one might see on an SE scsi bus.  Also, the drive hung
when I tried to run HP's ltt drive test, and when its cable was
unplugged a mysterious message appeared in /var/log "Transceiver state
has changed to LVD mode", implying that it was in some other mode
before.  Neither the cable nor the terminator the previous admin had
used was marked, but I figured the most likely thing was that the
terminator was SE and not LVD.  So an LVD terminator was ordered.

Today, with the new LVD terminator in place, the dd read/write tests
(from /dev/zero on write, to /dev/null on read) were much better (this
is with compression on):

    BS     N  WriteRate   ReadRate  (rates in MB/s)
 32768 20000       53.6       83.2
 65536 20000       97.2      145
131072 20000      122        167
262144 20000      134        183
524288 20000      149        192

mt -f /dev/nst0 compression off

apparently did what it said, even though neither "mt status" nor
the contents of "/sys/class/scsi_tape/nst0/*" changed.  With compression
off this table was generated:

    BS     N  WriteRate   ReadRate  (rates in MB/s)
 32768 10000       37        42
 65536 10000       49        54
131072 10000       54        57
262144 10000       57        58

Just under 60MB/sec seems to be the maximum tape transport read/write
limit.  Pretty reliably the first write from the beginning of tape was a
bit slower than writes started further into the tape.

I also tried tests at 524288, 1048576, and 2097152 block size.  The
the latter one unfortunately seems to have crashed the entire server.

Also, even with the right terminator, using HP's LTT device test (option
11) hung the tape drive so hard that a full power cycle on both the
drive and the server was needed to regain control.

The 60MB/sec is pretty impressive, much better than the 36MB/sec on our
SDLT 600.  On the other hand, the SDLT drive has never crashed our server! 


David Mathog
mathog at caltech.edu
Manager, Sequence Analysis Facility, Biology Division, Caltech

More information about the Beowulf mailing list