[Search for users]
[Overall Top Noters]
[List of all Conferences]
[Download this site]
| Title: | CSGUK_SYSTEMS |
| Notice: | No restrictions on keyword creation |
| Moderator: | KERNEL::ADAMS |
|
| Created: | Wed Mar 01 1989 |
| Last Modified: | Thu Nov 28 1996 |
| Last Successful Update: | Fri Jun 06 1997 |
| Number of topics: | 242 |
| Total number of notes: | 1855 |
82.0. "VMS bugs and problems" by KERNEL::ANTHONY () Mon Jan 08 1990 12:32
Hello All,
The following info was supplied by Rob Hasking...
Bottom line is advise upgrade to 5.2, or set mpw_iolimit to 1
Cheers, Brian
posted in csg_systems
---------------------------------------------------------------------------
This article details a problem in memory management on VMS V 5 systems prior
to VMS 5.2.
Resolution Statement.
As it has become evident, a possible page corruption under certain
conditions in any of VMS Version 5 releases PRIOR to V5.2 can cause a
system fatal bugcheck. Usually, the corrupted page is discovered
during an image or process run down, at which point the system
bugchecks. SOME of the discovered bugcheck reasons include WSLXVANMAT,
BADPAGFILD, PGFIPLHI, SSRVEXCEPT, INVEXCEPTN, with SOME of the addresses
on the Kernel stack pointing to MMG facility (typically MMG$DELPAG)
Following are some of the currently known characteristics that may
yield such system behavior, particularly in the environments with LARGE
amount of memory on some of the faster CPU's (uni OR multiprocessors):
o Large amount of paging activity can aggravate this condition;
o When a process needs to re-use a Working Set List Entry, a
dead page scan will get initiated. As a result, the most
likely candidates to be selected will be the WSLE's that
describe a page that is not valid (as indicated by its PTE)
but is still in the working set. This can cause increased
I/O activity as modified pages are forced to be written to
their backing store;
o The corrupted page is brought back in from the Page file, as
indicated in the system Crash Dump by the recently read IRP;
o Page file I/O activities are high and the number of outstan-
ding I/O's (controlled by MPW_IOLIMIT SYSGEN Parameter) is
set to higher than 1. (NOTE: the DEFAULT setting is 4)
An extensive cooperative effort was recently undertaken by the VMS
Engineering, CSSE and Field Customer Support organizations to better
understand the nature and specific causes of this system behavior.
A number of tests were executed on a range of various VMS software and
hardware configurations. As a result, this system behavior was
simulated on variety of high-speed CPU's and versions of VMS. While
the symptoms could be consistently reproduced on releases prior to
Version 5.2, the problem did not occur in an environment running VMS
V5.2. Furthermore, it was determined that none of the VMS V5 releases
experienced this behavior with MPW_IOLIMIT SYSGEN Parameter set to 1
(thus simulating pre-V5.0 behavior of the Modified Page Writer (MPW)
component of VMS Memory Management, MMG).
In addition, the problem was successfully simulated on a VMS V5.2
system ONLY when VMS code was modified NOT to include any of the
functional and design enhancements introduced in VMS MMG at V5.2 time
frame. Again, when the SYSGEN Parameter MPW_IOLIMIT was set to 1 on
the MODIFIED VMS V5.2 system, the problem would NO longer appear.
Therefore, based on the analysis of the above study, it is believed
that the following enhancements in VMS V5.2 MMG have significant
positive impact on the occurrence of this condition:
o Addition of selective page writing mechanism based on MPW's
internal states;
o Addition of termination logic to MMG$MPW_END routine such
that the Pending I/O, MPW state and Error conditions are
handled;
o Addition of threshold to check whether pending I/O requests
have been disabled.
The additional information that was obtained as a result of this
in-depth analysis was sufficient to better define the nature of this
system behavior and to identify a set of recommendations to systems
configurations experiencing this failure. Specifically, based on the
testing performed on various configurations, the analysis of the
results and of the V5.2 changes to VMS MMG, it was concluded that VMS
V5.2 significantly reduces the probability of experiencing this
problem, if not eliminates it all together.
Therefore, based on the results of this study, it is recommended that
systems experiencing this behavior are upgraded to VMS V5.2.
However, FOR THE INTERIM, those systems that continue to
run Pre-5.2 version of VMS AND are experiencing this problem
are strongly encouraged to reduce paging activities, increase
processes' working set sizes, and DEFINITELY set MPW_IOLIMIT SYSGEN
Parameter to 1. It should be noted that these actions are believed
NOT to have any significant impact on the overall system's
performance.
In addition to setting this parameter ,all VMS 5.1-n systems should
have the MMG patches installed to address other issues.
| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 82.1 | | KERNEL::MOUNTFORD | | Mon Feb 12 1990 03:29 | 64 |
| From: GOONS::EY8786U4 12-APR-1988 13:59:35.48
To: GOSSIP::WRIGHTON
CC:
Subj: from norm
From: GOONS::EY8786U3 12-APR-1988 12:08:19.11
To: EY8786U14,EY8786U4
CC:
Subj: sda help
From: KERNEL::PETTET "MNEMONIC NORM 11-Dec-1987 1224" 11-DEC-1987 13:31
To: @POST:RDCENG,QUINTANA
Subj: More SDA goodies.....
Gents,
Thought I would get in on the SDA mail coach,
typing :-
SDA> set proc/index=???
SDA> exam ctl$ag_clidata+2AF;200
Will exam the Command Recall Buffer. It will contain the last set of
commands that were typed on the terminal, it will not contain any commands that
were invoked by a DCL command procedure. The buffer is circular so if you want
to see all the commands typed then use:-
SDA> exam ctl$ag_clidata+2AF;405
The format of the displayed data is:-
1) Pointer (VA) to next command.
2) Null.
3) Size.
4) Command String.
5) Size
6) pointer......etc
Rat Holes......
===============
If you get the following the buffer is empty or you are pointing to the
wrong process!!!
SDA> exam ctl$ag_clidata+2AF;200
00000000 00000000 00000000 7FFE350F .5�............. 7FFE350B
Zeros suppressed from 7FFE351B through 7FFE370A
SDA>
You can try the exam ctl$ag...etc on the COMICS cluster (using your new
account) to test the procedure it does work and will work for VMS Versions 4.4
upwards.
Cheers...Norm
My thanks to Pete Dordoy TSC
|
| 82.2 | | KERNEL::MOUNTFORD | | Mon Feb 12 1990 03:29 | 21 |
| From: KERNEL::TRAVELL "John (I'd rather be windsurfing) Travell 09-Apr-1989 0826" 9-APR-1989 09:03:11.88
To: @POST:SYSTEMS
CC:
Subj: V5 crash command recall buffer
Hi all,
A snippet you have been waiting for, one of those things that could not
possibly do any sensible work without,... what's that!, you knew it?, could I
be the last to find out ??
The DCL command recall buffer MOVED in V5....
It is now at ... CTL$AG_CLIDATA+2B4;3F0
In symbolic terms, this has moved an enormous amount, all of 5 bytes...
However, the 24 bytes between CTL$AG_CLIDATA+29C & +2B4 appear to contain the
logical terminal name, eg. ._LOOKIN$VTA894.........
John Travell.
|
| 82.3 | Another way... | COMICS::EDMUNDS | | Sun Feb 25 1990 13:45 | 15 |
|
And further....
Another good way,sometimes possibly better,kindly passed on to
me by Ken Robb....
Pick up the process index
Find the term name if poss(sho proc/chan....LTAXXX etc)
sho dev LTAXXXX and look for the UCB address (<ADDRESS>)
read sys$system:sysdef
for <ADDRESS> /type=UCB (look for UCB$L_TT_TYPAHD (<ADDRESS2>)
ex <ADDRESS2> ;200(or so...)
|
| 82.4 | | KERNEL::MOUNTFORD | | Tue Jul 03 1990 12:37 | 151 |
|
From Geoff Judd.
A new BACKUP patch has been released which fixes a number of problems.
The problems are listed in the .README file which is included below.
This is the patch which is referred to on STARS as CSCPAT_0101 and
says that its VOID name is BACKUP$PATCH01_531.
The patch is called BCK$IMG01_531 and is note #142 in the VMS_PATCHES
notes conference.
Geoff Judd.
Patch Identification:
BCK$IMG01_531
Patch Policy Statement:
This image is to be used ONLY as part of a solution to a specific
problem and NOT as a troubleshooting method, i. e. for 'quick fix
tries' at a customer site.
Image Description:
The Backup image provided in this kit is built based on VMS V5.3-1.
The Backup cumulative image can be applied to any VMS system running
VMS V5.2 to V5.3-1, inclusive. The following is a list of bug fixes
in this image. A brief description of the problem or symptom is
included along with which version(s) these problems have been
reported on.
Problem Descriptions:
- Backup performance degrading on non-CRC emulated machines
V5.2 - V5.3-1
As of v5.2, any machine that does not do CRC emulation is probably
seeing a performance degredation if they are using the /CRC qualifier
( default, which is recommended by Digital ). Machines that are
likely to see this problem are 11/750, 11/780 and 11/785.
- ODS-1 data integrity issue
V5.2 - V5.3-1
The problem's symptom was when performing a /VERIFY on the save
operation, Backup VBNMISSING errors were reported. The problem
described is more apparent on fragmented disk structures. Block
count for each header extent was being calculated improperly. This
in turn caused the file's size to be truncated and compromised data
integrity.
- ODS-1 Backup looping problem
V5.2 - V5.3-1
When performing a Backup operation on an ODS-1 disk structure,
the Backup would hang. This can be detected by the ^T function,
and the amount of time the Backup operation takes. The problem
described is more apparent on fragmented disk structures. This
is usually seen on large files with multiple header extents.
- BACKUP/DELETE may erroneously delete files
V5.3 - V5.3-1
A input file was erroneously deleted when an error was encountered
creating the output file during a disk to disk image backup or backup
copy operation.
- Reserved operand when mounting second tape volume
V5.3 - V5.3-1
When Backup issued the command to mount the second volume, a
reserved operand error was likely to occur. This was seen when
specifying /NOASSIST on the Backup command.
- ACCVIO at PC F9AE
V5.2 - V5.3-1
An access violation occurred at the PC F9AE in Backup.
- Backup hanging when using /COMPARE
V5.2 - V5.3-1
When doing a Backup compare operation, the Backup would hang if
process quotas are below the recommended values. See section 8.2
in the "VMS System Manager's Manual" for Digital recommended values.
- Restoring of multi-volume disk savesets
V5.2 - V5.3-1
During the restore of a multi-volume disk saveset and specifying
/ASSIST ( default ), Backup would get into an endless loop,
requesting the user to mount the second volume and never returning
control back to the user.
Installation Instructions:
This update is designed to be installed with the VMSINSTAL utility
as follows:
1. Ensure that the input save set is located on disk as opposed to
magnetic tape.
2. If this kit is received on a magnetic tape, then MOUNT the tape
files-11, and COPY its contents to disk prior to issuing the
following command -
@SYS$UPDATE:VMSINSTAL BCK$IMG01_531 [ location of .A file ]
Limiting Parameters:
None.
Special Notes:
Note that this is an image update. Any patches applied to BACKUP
prior to this update will be lost.
DCLTABLES is modified with this kit to include qualifiers added
in V5.3-1. Users will have to log out and log back in to incorporate
this new version of DCLTABLES.
Listing of save set(s)
Save set: BAC531.A
Written by: HELLWIG
UIC: [000011,002020]
Date: 11-MAY-1990 10:41:15.62
Command: BACKUP/NOASSIST/COMMENT=VAX/VMS SPKITBLD Procedure/INTER/LOG/VERIFY *.COM,*.CLD,BACKUP.EXE;6 SODAK$DKA0:[0,0]BAC531.A/LABEL=(BAC)/SAVE/BLOCK=9000/GROUP=25
Operating system: VAX/VMS version V5.3
BACKUP version: V5.3
CPU ID register: 0A000004
Node name: _SODAK::
Written on: _SODAK$DKA0:
Block size: 9216
Group size: 25
Buffer count: 14
[HELLWIG.BUILD_531]KITINSTAL.COM;8 7 11-MAY-1990 10:40
[HELLWIG.BUILD_531]BACKUP.CLD;1 8 8-NOV-1989 16:02
[HELLWIG.BUILD_531]BACKUP.EXE;6 221 24-APR-1990 16:04
Total of 3 files, 236 blocks
End of save set
|