FASTSKIP not working on iVMS V8.3-1H1???

[Veli Körkkö]

Customer Alpha systems had V7.3-2 and pretty outdated patches. ABS v4.5 however started backups quite quickly, i.e. from BATCH JOB START to real work only a few minutes.

On Integrity (VMS V8.3-1H1, latest patches) it seems that the SKIP TO END OF TAPE takes quite long time.

Tape drive is MSL6060 with E1200-320, same device for both clusters.


My hunch would be that on ALPHA ABS/VMS did perform FASTSKIP to end of tape and on Integrity for whatever reason ABS/VMS might do much slower SKIP BY RECORD

Anybody having same problem?

[Volker Halle · Accepted solution]

Veli,

last time I've looked, BACKUP itself uses a call SKIP_TM(32767). I would assume this to be the equivalent of SET MAG/SKIP=FILES:32767

Did you test this operation explicitly on both architectures ?

Volker.

[Volker Halle]

Veli,

if you look at the tape from SDA, while such a 'skip' operation is underway, you may see the current IO function and the modifiers:

$ ANAL/SYS
SDA> SHOW DEV $2$MGAx
...

Volker.

[Wim Van den Wyngaert]

HELP set mag/fast.

Or try /fast=always.

Wim

[Veli Körkkö]

The actual tool used here is ABS, V4.5. On the beginning of a "backup sequence" ABS does quite a few things. Most notably it mounts the tape and skips to the end of tape.

It would assume that it does SKIP BY FILE as it "needs to count" the filemarks so it is able to put that informantion into history file along with saveset name etc stuff.

After ABS(?) has reached "end of tape" it will actually start the real backups as subprocesses.

And now it seem that ABS on Integrity spends every day longer and longer before we get to real work.

I already suggested customer $SET MAGETAPE/SKIP=ALWAYS $2$MGAx:

Also I was planning as a test

$ set noon
$ mount/fore/nounload 'device'
$ show time
$ set magt/skip=end 'device'
$ show time
$ dismount/nounload 'device'

to be done from both Alpha and Inegrity against same tape libary/tape drive/tape

so as to see whether there would some major
difference between a ES40 running V7.3-2
and RX6600 running V8.3-1-1H1 (with pretty much uptodate patches)


_veli

[Volker Halle]

Veli,

did you check for a ABS_NO_FAST_SKIP logical (see Guide to Operations chapter 12.1.5) Ã

Volker.

[Veli Körkkö]

not defined. would have to be defined intentionally. was not.

well, takes some time before I get some hard facts since this is a customer system.

_veli

[Veli Körkkö]

Well, there was no noticeable delay today after $SET MAGTAPE/FAST=ALWAYS had been used yesterday.

Customer tests indicate that

$ mount/fore/nounload 'device'
$ set magtape/skip=end 'device'

takes much longer when $SET MAGTAPE/FAST=PER_IO
compared to $SET MAGTAPE/FAST=ALWAYS.

However this behaviour was observed both on cu old ALphas and new Integritys.

The underlying problem is not solved really but at least we appear to have now a intermediate workaround for the issue.

thanks everybody

[Wim Van den Wyngaert]

I already noticed that the set magtape is remembered after the dismount. I guess it's changing a device setting, not the current tape operation.

Wim

[Volker Halle]

Veli,

this may be a question of defaults ...

If you SET MAG/FAST=PER_IO (or if this is the default setting), each IO$_SKIPFILE QIO would require the IO$M_ALLOWFAST modifier for fast skip operation. SET MAG/SKIP=END does use a simple IO$_SKIPFILE QIO with P1 = 32767, so it would NOT honour the /FAST=PER_IO setting.

SET MAG/FAST=ALWAYS would always try to use the fast method.

A SHOW DEVIVE/FULL tape should tell you the fastskip settings as

'device supports fastskip' with an additional (always) or (disabled) or (per_io)

These settings are in UCB$L_DEVDEPND2 - visible with SDA.

Volker.