[comp.os.vms] Experiences with Digidata's Gigastore

BART@HDETUD53.BITNET (Bart Zorn) (01/18/88)

We have tested the Digidata Gigastore unit on the following
configuration:

        VAX11/750, FPA, 8Mbyte memory,
        CI750, HSC50, 2 * RA81, 1 * RA60
        Emulex TC7000, Cipher M990
        DEUNA, DMF32, RL211, 2 * RL02, LP11
        UB NIU-DMF, Excelan EXOS 204 TCP/IP controller.

When used with the Dilog tapecoupler wich was delivered with the
video tape unit, an UNIbus interface which emulates the DEC TS11
tapeunit, and the also supplied modified TSDRIVER.EXE the
subsystem performed very well. This driver is installed using
VMSINSTAL. Glancing throug KITINSTAL.COM learned that Digidata
expects this driver to break with VMS V5.x!

The first experiment was a repeated backup of the RA60 disk which
was 85% allocated (170 Mb). The contents of the disk fitted ten
times on one E-180 VHS cartridge. Average time needed for one
backup was 35 minutes, i.e. 290 Mb per hour. BACKUP qualifiers
used were: /IMAGE/BUFF=5/NOCRC/GROUP=0/BLOCK=32768. This test
was done during night hours with a lightly loaded system.

The second experiment was a full restore of one of the created
backup's from the previous test. During normal daily operations
this took 90 minutes, almost three times as much as the
correspondig backup. No errors were signaled. The video unit had
to do a short rewind to reposition the tape quite a lot of times,
which will probably explain the extra time needed.

The third experiment was a nigtly backup of all three diskunits.
The following is an extract from the logfile:

$ SET NOON
$ SHOW TIME
  14-JAN-1988 23:26:26
$ ALLOCATE MSA0:
%DCL-I-ALLOC, _TUDEDV$MSA0: allocated
$ INIT MSA0: TEST03
$ SHOW TIME
  14-JAN-1988 23:32:34
$ MOUNT/FOR/NOASSIST MSA0:
%MOUNT-I-MOUNTED, TEST03 mounted on _TUDEDV$MSA0:
$ SHOW TIME
  14-JAN-1988 23:33:15
$ BACKUP/IMAGE/BUFF=5/FAST $1$DUA0: -
      MSA0:DUA0.BCK/NOCRC/GROUP=0/BLOCK=32768
$ SHOW TIME
  15-JAN-1988 00:56:00
$ BACKUP/IMAGE/BUFF=5/FAST $1$DJA1: -
      MSA0:DJA1.BCK/NOCRC/GROUP=0/BLOCK=32768
$ SHOW TIME
  15-JAN-1988 01:29:48
$ BACKUP/IMAGE/BUFF=5/FAST $1$DUA2: -
      MSA0:DUA1.BCK/NOCRC/GROUP=0/BLOCK=32768
$ SHOW TIME
  15-JAN-1988 01:53:58
$ SET MAGTAPE/REWIND MSA0:
$ SHOW TIME
  15-JAN-1988 01:55:53
$ BACKUP/LIST=NLA0: MSA0:*.*
$ SHOW TIME
  15-JAN-1988 04:09:54
$ SET MAGTAPE/REWIND MSA0:
$ SHOW TIME
  15-JAN-1988 04:11:24
$ DISMOUNT/UNLOAD MSA0:
$ DEALLOCATE MSA0:
$ EXIT
  SYSTEM       job terminated at 15-JAN-1988 04:11:26.07
  Accounting information:
  Buffered I/O count:        51027      Peak working set size:   919
  Direct I/O count:         121215      Peak page file size:    1428
  Page faults:                1314      Mounted volumes:           1
  Charged CPU time:     0 00:31:16.43   Elapsed time:     0 04:45:05.01


Disk statitics were:

                  Free     Allocated    Total    Percent
  Disk           blocks     blocks      blocks  allocated
  ----           ------    ---------    ------  ---------
$1$DUA0:         214440      676632     891072     75.9
$1$DJA1:          68634      331542     400176     82.8
$1$DUA2:         604587      286485     891072     32.1
                 ======      ======    =======    =====
                 887661     1294659    2182320     59.3


We used BACKUP/LIST as verification as recommended by Digidata
which also used almost twice as much time as the corresponding
backup. I did not try a BACKUP/VERIFY because this would rather
useless on a running system anyway.

Conclusion: the Gigastore unit provides an excellent solution to
the growing problem of backing up several hundreds of megabytes.
Speed and capacity of the unit are very good. We will order one!

We tried to connect the Gigastore to our Emulex TC7000 tape
coupler, which emulates a massbus adapter and uses the pertec
interface to the tapedrive(s). This combination wouldn't work.
Either the TC7000 didn't see the Gigastore (configured as unit
#1), or the 750 didn't see the TC7000 at all. We decided not to
bother and use the TS11 emulation, which suits our requirements
very well.
Maybe a real "interface-guru" can supply some hints as to what
could be wrong?


Delft, 18-Jan-1988

Bart Zorn       (BART@HDETUD53.BITNET)
Delft University of Technology
Faculty of Electrical Engineering
P.O. Box 5031
NL - 2600 GA  Delft
Phone: +31 15 782894