| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 345.1 | Subprocess is not idle ? | UTRTSC::SCHOLLAERT | Half Dutch - Half Belgium | Tue Mar 31 1992 15:53 | 12 |
| > My customer said that the new memory management from VMS 5.4-3
> doesn't work for ALL-IN-1 subprocesses . In other words , VMS is
> not swapping out ALL-IN-1 subprocesses , even if they're doing
> nothing .
VMS works like expected. IDLE processes are swapped out.
$ sho proc /cont shows that the ALL-IN-1 subprocess is spending
CPU time, even when the main process is swapped.
Don't ask me why...
|
| 345.2 | spending CPU time | VAXSPO::ANDRE | | Tue Mar 31 1992 20:29 | 8 |
|
You're right , the subprocess is spending CPU time , but how can
I explain that to my customer ???
Do you know any reason that forces subprocesses to spend CPU time ?
Thanks ,
Andre'
|
| 345.3 | | SHALOT::LAMPSON | Whatever ya do, ya gotta have FUN! | Tue Mar 31 1992 22:34 | 8 |
| Well, every time a permanent symbol in ALL-IN-1 is created or
updated and if the subprocess is open, a corresponding DCL symbol
is created or updated.
For example, if the code GET $MIKE = "ANYTHING HERE" is executed,
the DCL symbol MIKE is created.
_Mike
|
| 345.4 | Odd | AERO::TALLETT | Just one more fix, then we can ship... | Wed Apr 01 1992 09:11 | 12 |
|
RE: .3
How can permanent symbols be getting created if the main process
is swapped?
This sounds wierd to me. Does it use LOTS of CPU or just the odd
tick? Its probably not lots or we would have had it bugged. So
whats going on? Something to do with mailboxes?
Regards,
Paul
|
| 345.5 | Slowly Ticking | UTRTSC::SCHOLLAERT | Half Dutch - Half Belgium | Wed Apr 01 1992 09:40 | 34 |
| RE. 4
> This sounds wierd to me. Does it use LOTS of CPU or just the odd
> tick? Its probably not lots or we would have had it bugged. So
On a 6000, the process uses in 5 elapsed minutes:
* 0.1 of a second CPU
* 100 page faults
* no I/O
Hereby some snapshots of my batchjob checking the subprocess...
Elapsed CPU time: 0 00:00:02.76
Connect time: 0 00:50:28.43
Elapsed CPU time: 0 00:00:02.77
Connect time: 0 00:50:38.69
Elapsed CPU time: 0 00:00:02.78
Connect time: 0 00:51:09.32
Elapsed CPU time: 0 00:00:02.79
Connect time: 0 00:52:10.58
Elapsed CPU time: 0 00:00:02.80
Connect time: 0 00:52:20.93
Elapsed CPU time: 0 00:00:02.81
Connect time: 0 00:52:41.46
Elapsed CPU time: 0 00:00:02.82
Connect time: 0 00:52:51.67
|
| 345.6 | Einstein says you can't watch without disturbing | IOSG::SHOVE | Dave Shove -- REO-D/3C | Wed Apr 01 1992 12:33 | 10 |
| I expect you know this, but others may not -
monitoring a process with SHOW PROC actually causes it to use a little
CPU (because it's done using inter-process ASTs or something - it's a
long time since I went on a VMS Internals course :-)). However, I don't
believe it uses page faults, except sometimes for the first time.
So I agree with Paul - it does sound wierd.
Dave.
|
| 345.7 | Einstein : candidate for the VMS Internals course | UTRTSC::SCHOLLAERT | Half Dutch - Half Belgium | Wed Apr 01 1992 14:31 | 43 |
| Dave,
> I expect you know this, but others may not -
>monitoring a process with SHOW PROC actually causes it to use a little
>CPU (because it's done using inter-process ASTs or something - it's a
No, I didn't. Don't tell my boss. But.....
> -< Einstein says you can't watch without disturbing >-
Einstein must have lived long before "$SHOW SYSTEM" was invented.
Have a look at the following 30 second snapshots from
my main process (ALL-IN-1) and its subprocess (A1_SUB).
VAX/VMS V5.5 on node UTRYIT 1-APR-1992 13:41:51.46 Uptime 1 21:02:15
Pid Process Name State Pri I/O CPU Page flts Ph.Mem
2CE01B76 A1_SUB LEF 6 121 0 00:00:01.83 665 108 S
2CE01B76 A1_SUB LEF 6 121 0 00:00:01.83 665 108 S
2CE01B76 A1_SUB LEF 6 121 0 00:00:01.83 665 96 S
2CE01B76 A1_SUB LEF 6 121 0 00:00:01.83 680 108 S
2CE01B76 A1_SUB LEF 5 121 0 00:00:01.84 680 96 S
2CE01B76 A1_SUB LEF 6 121 0 00:00:01.85 695 108 S
2CE01B76 A1_SUB LEF 6 121 0 00:00:01.85 695 108 S
2CE01B76 A1_SUB LEF 6 121 0 00:00:01.85 695 96 S
2CE01B76 A1_SUB LEF 6 121 0 00:00:01.86 710 108 S
2CE01B76 A1_SUB LEF 6 121 0 00:00:01.86 710 108 S
2CE01B76 A1_SUB LEF 6 121 0 00:00:01.87 710 96 S
2CE01B76 A1_SUB LEF 6 121 0 00:00:01.87 710 96 S
2CE0027E ALL-IN-1 LEF 4 3246 0 00:00:45.13 16277 2265
2CE0027E ALL-IN-1 LEF 4 3246 0 00:00:45.13 16277 2265
2CE0027E ALL-IN-1 LEF 4 3246 0 00:00:45.13 16277 2265
2CE0027E ALL-IN-1 LEF 4 3246 0 00:00:45.13 16277 2265
2CE0027E ALL-IN-1 LEF 4 3246 0 00:00:45.13 16277 2265
2CE0027E ALL-IN-1 LEF 4 3246 0 00:00:45.13 16277 2265
2CE0027E ALL-IN-1 LEF 4 3246 0 00:00:45.13 16277 2265
2CE0027E ALL-IN-1 LEF 4 3246 0 00:00:45.13 16277 2265
2CE0027E ALL-IN-1 LEFO 4 -- swapped out -- 1886
2CE0027E ALL-IN-1 LEFO 4 -- swapped out -- 1886
2CE0027E ALL-IN-1 LEFO 4 -- swapped out -- 1886
2CE0027E ALL-IN-1 LEFO 4 -- swapped out -- 1886
|
| 345.8 | The elevated priority is a big clue (for what?) | BUFFER::VICKERS | If it helps a customer, DO IT | Wed Apr 01 1992 17:35 | 26 |
| Dave, as always, is right on the money. Of course, it was an physicist
named Heisenburg who developed the theory of uncertainty which states
that the act of measuring an event affects the event. As explained
quite fully on page 333 of the VAX/VMS INTERNALS AND DATA STRUCTURES by
Ruth Goldenberg, $GETJPI must use a Special Kernel Mode AST to obtain
process information since that information is maintained within the
process space. Sounds like Object Oriented Programming doesn't it?
In any case, the CPU and page fault behavior is consistent with the
effect of the ASTs being processed by the processes being observed.
However, the changes in priority and memory usage indicate that the
subprocess is doing other work, especially IO (causing the priority to
be elevated).
It is also interesting that the page faults for the subprocess jump by
15 at roughly the same point as the main process was swapped out.
I assume that you have done a SHOW PROC/CONT on the sub process to be
sure that there are no images running there? Of course, DCL itself is
more than capable of doing enough IO to keep the priority elevated.
Is the job tree complicated by having sub processes under A1_SUB by any
chance?
The current evidence is most strange, indeed,
don
|
| 345.9 | The Morning after | UTRTSC::SCHOLLAERT | Half Dutch - Half Belgium | Thu Apr 02 1992 09:26 | 47 |
| Hello,
> In any case, the CPU and page fault behavior is consistent with the
> effect of the ASTs being processed by the processes being observed.
> However, the changes in priority and memory usage indicate that the
> subprocess is doing other work, especially IO (causing the priority to
> be elevated).
I think this is correct for $SHOW PROCESS.
$SHOW SYSTEM does not affect any in or outswapped process. Don't
ask my why. To avoid the watcher-bewatched conflict I went home
and had a good night sleep.....
Last result of SHOW SYSTEM on 2 apr:
2CE01B76 A1_SUB LEF 6 121 0 00:00:01.87 710 96 S
SHOW SYSTEM on 3 apr on the morning after
2CE01B76 A1_SUB LEF 4 121 0 00:00:05.86 13124 93 S
Main process still swapped out...
2CE0027E ALL-IN-1 LEFO 4 -- swapped out -- 1886
> It is also interesting that the page faults for the subprocess jump by
> 15 at roughly the same point as the main process was swapped out.
After the main process is swapped out, the subprocess keeps paging...
> I assume that you have done a SHOW PROC/CONT on the sub process to be
> sure that there are no images running there? Of course, DCL itself is
> more than capable of doing enough IO to keep the priority elevated.
Yip. The only thing I did in the subprocess was $set proc/name=A1_SUB.
> Is the job tree complicated by having sub processes under A1_SUB by any
> chance?
No tree.
> The current evidence is most strange, indeed,
Indeed. Check on VMS 5.3, the subprocess is not using any resources...
Should it be VMS 5.4 forcing the subprocess to adjust its working set ?
Whats so special about the ALL-IN-1 subprocess ?
|
| 345.10 | Any chance there's an old VTX image in there?? | IOSG::SHOVE | Dave Shove -- REO-D/3C | Thu Apr 02 1992 12:53 | 12 |
| That's A LOT of page-faulting. There obviously really is something
going on in the subprocess (or $SHOW SYSTEM is a liar :-)).
The main process will sometimes use CPU when theoretically idle if
broadcast trapping is turned on and there are some broadcasts (new
mail, operator does REPLY/ALL, etc). Is there any chance that there's
some image lying around in the subprocess that also does broadcast
trapping (for example, VTX, TPU)?? Indeed, is there any image in the
subprocess (according to SHOW PROC/CONT or f$getjpi(pid,"IMAGNAME"))?
All very curious . . .
Dave.
|
| 345.11 | story continues | UTRTSC::SCHOLLAERT | Half Dutch - Half Belgium | Thu Apr 02 1992 17:26 | 14 |
| Dave,
There is nothing running in my subprocess.
I just invoke ALL-IN-1 and $set proc/name=a1_sub
I found the cause of the problem. Its the new memory
managament mechanism ITSELF !
Set MMG_CTLFLAGS to 0 and no pagefault and no cpu-tick will
occur.
Big question: whats so special about the ALL-IN-1 subprocess ?
Jan
|
| 345.12 | doing something? | IOSG::DAVIS | Mark Davis | Thu Apr 02 1992 18:20 | 25 |
|
Hi,
I have looked at this on a 5.3 system - The ALL-IN-1 subprocess seems
to have a timer queue entry that is scheduled every minute exactly. The
timer queue entry appears immediately after ALL-IN-1 creates the
subprocess - this is stepping through with the debugger.
On a 5.3 system the only noticeable thing that occurs when the timer
request goes off is that the process gets a priority boost.
However on a 5.5 system with memory management I guess that the
following happens. Because of small activity every minute, the process
is identified as an periodically waking process - i.e one that is
liable not for swapping but for trimming of 25% of its pages when it
enters the wait state (assuming that free memory has fallen to less
than twice FREEGOAL). The next time it wakes up it needs to bring those
pages into memory - thus causing pagefaults and CPU usage. The CPU
would probably be too small to be detectable without the page faults.
What it is trying to do every minute I don't know.
Mark
|
| 345.13 | sorry , but ... | VAXSPO::ANDRE | | Thu Apr 02 1992 21:19 | 5 |
|
Isn't possible that the ALL-IN-1 subprocess is created with a flag
that disables swapping ?
Andre'
|
| 345.14 | Not that | IOSG::DAVIS | Mark Davis | Fri Apr 03 1992 12:10 | 20 |
|
No, unless disable swapping is the default.
There seem to be two main problems here with the ALL-IN-1 subprocess.
The first, as was said earlier, is that whenever a permanent symbol is
created or modified in the main process, then the corresponding symbol
is created or modified in the subprocess. This is done in the
subprocess. Permanent symbols are frequently modified in the main
process and so there will be continual activity in the subprocess
while the main process is active.
The other problem is that even when the main process is inactive or
swapped out, there is a small amount of activity in the subprocess
that is activated by an entry in the timer queue every minute. Perhaps
this is some side effect of the create process system service.
Mark
|
| 345.15 | No timed wait that I can see; no flags either | IOSG::SHUV::IOSG::SHOVE | REO/D-3C | Fri Apr 03 1992 16:56 | 8 |
| No. ALL-IN-1 issues a very simple $CREPRC call, with no flags or anything
(other than pointing the input and output at appropriate mailboxes.)
I can't find anything in the code which waits for a minute.
Strange . . .
Dave.
|
| 345.16 | Effected more than ones a minute | UTRTSC::SCHOLLAERT | Half Dutch - Half Belgium | Fri Apr 03 1992 17:28 | 14 |
| > I can't find anything in the code which waits for a minute.
The subprocess is more than ones a minute effected.
Have a look at pagefaults during the following 10 sec snapshots.
Pid Process Name State Pri I/O CPU Page flts Ph.Mem
2CE02764 A1_SUB LEF 7 35 0 00:00:00.75 364 97 S
2CE02764 A1_SUB LEF 7 35 0 00:00:00.75 364 97 S
2CE02764 A1_SUB LEF 6 35 0 00:00:00.75 366 95 S
2CE02764 A1_SUB LEF 6 35 0 00:00:00.75 379 108 S
2CE02764 A1_SUB LEF 6 35 0 00:00:00.75 379 108 S
|