Thanks to the following for their swift responses:
Renaud KERIVEN <keriven_at_corea.enpc.fr>
Hellebo Knut <Knut.Hellebo_at_nho.hydro.com>
Arne Steinkamm <arne_at_Steinkamm.COM>
qu_at_madrad.radiology.wisc.edu (Jane Qu )
Malcolm Dunnett <dunnett_at_mala.bc.ca>
Tim Mooney <mooney_at_dogbert.cc.ndsu.NoDak.edu>
Everyone agreed that switches 1 and 2 control compression as follows:
SWITCH 1 = 1 means the drive defaults to writing compressed data
= 0 means the drive defaults to writing non-compressed data
SWITCH 2 = 1 means the host has the ability to turn compression on/off
= 0 means the drive ignores host requests to change compression
The meaning of switches 3-8 changed with firmware release 9.47 (or
9.49?). Tim Mooney sent me a posting from comp.sys.hp.hardware which
describes everything. Apparently: "there is no 1-1 mapping of features
to switches. A specific pattern of switches will enable a SET OF
FEATURES". Having quoted that, it seems that switch 3 controls MRS,
switch 4 controls "No Dataphase disconnect" and switch 6 "Performance"
when switches 5,7,8 are set to 1-10.
Using "scu -f <device> show device" tells me that the revision level for
our device is 9503, so for a DEC host I think the following settings
apply:
MRS No Data
ON Phase Disconn Performance 345678
111110
1 011110
1 111010
1 1 011010 ('recommended')
1 1 101010
1 1 1 001010
Of these, Malcolm Dunnett reports that 011010 are "the recommended
settings for VMS, Ultrix and DUnix". The meaning of the features are as
follows:
MRS on
------------
TRUE: MRS (Media recognition system) is implemented.
Tapes have to be of DDS grade, otherwise they are treated
as write protected and write commands will be rejected
with sense key=7, additional sense =3000h.
FALSE: MRS is not implemented. You can write any tape you like.
no data disconnect
-----------------------
TRUE: No disconnect during data phase.
NB: save data pointers may still be sent.
FALSE: Disconnects may occur during data phase.
Comments: Useful for hosts which cannot disconnect on ODD BYTE
boundaries. Some SGI hosts can't disconnect on ODD
WORD (32 bit) boundaries so this feature should be
enabled for SGI connect.
Downside = DOn't do SCSI operations exceeding 64Kbytes
otherwise the bus could hang-up waiting for device-side
I/O. The drive can normally guarantee to have at least
64K of data/space before it starts data phase.
performance
-----------------
TRUE: ``immediate'' and ``inifinite flush'' are selected (TRUE)
FALSE: ``immediate'' and ``inifinite flush'' are not selected (FALSE)
Comments: This can dramatically improove performance when applications
"do silly things", like try and disable buffering or write
lots of non-immediate filemarks. If they do silly things and
this feature is not enabled then the drive will spend most
of its life stream-failing.
(inf.fl.) Infinite flush means data is left lying around in the
buffer for long periods of time (and is subject to LOSS
if power fails). The plus side of this is that VERY
SLOW applications don't continually stop/start the
mechanism (it will flush every 5 seconds by default).
ANother plus point of this feature is that you won't
loose capacity if the drive flushes partially written
groups.
(immed.) SCSI commands affected include: Load, Rewind, Erase,
Write Filemark. Assuming immediate mode for filemarks
can drastically improove the performance of systems which
fail to set this bit when sending write filemark commands.
The downside is that you don't flush data to tape in
response to a filemark command, and for some applications
this compromises data-integrity checkpointing.
Thanks and regards,
Jon
[ Jonathan.Buchanan_at_ska.com ]
Received on Tue Apr 02 1996 - 11:00:43 NZST