| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 106.1 | Module 11: Performance Management Posted for Review | WHEEL::MOSTEIKA | Paul @ GSF/B21, DTN 264 (884)-2832 | Wed Sep 25 1991 17:03 | 16 |
| Module 11 - Performance Management, is ready for review.
Please check for technical content, organization and sequencing.
The Performance module is available on:
SUPER::ES$REVIEW:[SYSNET_III]SYSNETIII_CHAP11.PS
Regards,
Paul M.
|
| 106.2 | Again: this is not task oriented. | NWGEDU::RODENBURG | Ed. Services, The Netherlands | Tue Oct 01 1991 13:00 | 42 |
| Hello world,
I copied the chapter, and went through the material.
I am realy astonished about the number of studentpages: 150
I cannot imagine, that this is only 1 (one) chapter in the SYSNET III
course. What is the total courselength?
Some general comments on it:
- in my opinion it is going too far for a student of the intendeded
audience.
As an example, why talking talk about Network performance in
terms of line utilization, all sorts of NCP counters, rated
throughput, real throughput etc., etc.
- in my opinion it is too much for one chapter in a course with a
reasonable length
- in my opinion the information in this chapter is NOT usefull for
the student of the inteded audience in this course
I will not go into detail, reviewing this material, because I don't
know where to start. Most important: the starting point, the setup of
the chapter is totally wrong.
This is NOT task-oriented and is not what we need!!
In my opinion we must first go back, and define what the student of
the intended audience must be able to do after complemention of this
module.
Second, the developer must think about the way how the subjects must be
presented in a TASK-ORIENTED way.
Sure it is NOT: taking pieces from existing courses together and present
it as a new chapter.
I am sorry, to come back again and again on the same item: a
task-oriented course string will need a different way of presenting the
information.
Regards,
Joop
|
| 106.3 | Not suitable !! | MINDER::GRAVESG | Geoff Graves,EDU(UK); DTN 851 2637 | Wed Oct 02 1991 06:38 | 45 |
|
I have difficulty in believing what I see in Chapter 11 !!.
Page 1-3a says "Lecture time is approximately 2 hours..."
Having taught the VMS System Performance Management course (9767)
over 20 times, mainly to customers, I reckon that 2 *days* would barely
be enough time to cover this, this _plethora_ of material! And that
would not leave any time for labs or students' questions/discussions!!
And why suggest 3 hours for lab work that normally takes about 30/45
minutes?
This is a totally unacceptable amount of detail for SYSNET III.
If it's left as it is, I can see that all instructors will *have* to leave
large chunks of it out, to fit everything else in the course.
That, obviously, is going to be bad for the student, the instructor,
EduServices and Digital!
I can't help wondering: has the developer ever stood up and taught this
material? In the classroom situation, there has to be absorption time.
Students have to be (at least, partly) comfortable with a topic before
moving on to the next related topic. The instructor may have to
explain a point 2 or 3 times for some students and satisfy themselves
that the class is with them, before proceeding. I'd have difficulty
just *reading* the 150 pages in 2 hours, never mind explaining and
clarifying points for a mixed bag of students as well !
Like Joop in .2, I also have not bothered reviewing the material in
detail as it is so completely unsuitable for this course. Just lifting
large chunks of material from existing courses (VMS Perf, DECnet
Management) and adding one or two new pages, is *not* what's needed
here.
The specification of SYSNET III contains the following entry under
section 1.9 Nongoals...
"System performance management and tuning beyond some introductory
tasks (taught in the VMS System Performance Management course)"
What has gone wrong here ???
|
| 106.4 | exit | NWGEDU::HOMPUS | | Wed Oct 02 1991 09:00 | 46 |
|
Colleague,
I'm still very impressed about chapter 11 of SYSNET III.
This is almost as much as the performance course handout.
I have given that course 40 times, and I'm always short
of time. Why? We have incorporated lab exercises in that
training because in the beginning we had complains the
course was too theoretical. I spend about 70% : 30%
lecture :lab in this course, and the student still want
more cases to analyze.
Back to Chapter 11 of SYSNET III.
How much time is available to teach chapter 11, it has about
150 student pages. On page 1-3a is mentioned 2 hours lecture and
3 hours lab. I think this is absolutly not realistic.
I do not recognize the TASK ORIENTATION in this module, and
is that not the fundament of SYSNET I,II and III.
What do we need, if we want to have it TASK ORIENTED.
Well we should go back to the detail design of SYSNET I,II and III,
there we must find the definition of the goals for these three courses.
What the students want ?
- a good understanding of sysgen
- " " " autogen
- Why and how modparams.dat
- What does Monitor show and how do I have to read it.
- How to work with Install
- Several Show commands.
- Some RMS commands about: Analyze/rms, Convert, Edit/FDL
- Accounting utility
Regards,
Piet Hompus.
Educational Services
Nieuwegein
The Netherlands.
|
| 106.5 | | SUPER::MATTHEWS | | Wed Oct 02 1991 14:35 | 2 |
| Revision to this chapter is in progress. Stay tuned.
|
| 106.6 | Asking for a newer version of this moduel | NWGEDU::RODENBURG | Ed. Services, The Netherlands | Fri Nov 08 1991 10:48 | 7 |
|
I want to review the material I received, but I realy still don't like
to go through the 117 pages of this chapter. Is there a more recent
version of the chapter included into the presented complete
courseguide or do you expect us to specify what has to be included?
Joop
|
| 106.7 | Revised Performance Module | HARDY::MOSTEIKA | Paul the Nomad Developer | Mon Nov 18 1991 17:06 | 24 |
| The Performance module has been modified a few times, both before and
after the Pilot. Sorry for the delay, but it made more sense to hold it
back during all the changes.
Concepts were trimmed back. The Network Performance info was put in an
appendix, then eliminated all together. The page count is down around
60 - some pages by trimming some of the superfluous. And yes, the
lecture time and lab time allocated is 3 hours each. The main topics
are:
Intro to Performance
Controlling Memory Resources
" I/O "
" CPU "
Thanks in advance for any feedback. The Monitoring module
should be ready soon.
The .PS file is available on:
SUPER::ES$REVIEW:[SYSNET_III]SYSNETIII_CHAP10.PS
Regards,
Paul M.
|
| 106.8 | | NWGEDU::HOMPUS | | Fri Nov 22 1991 02:33 | 334 |
| Hello,
I have reviewed chapter 10 of SYSNET III.
Some general comments:
I still do believe it is to large for SYSNET III.
I disagree with the suggested time, in mine opinion it is not possible
to cover this in 3 hours.
I think that the level of depth is to great, and if we stay at this level
we must be more accurate.
In this version of chapter 10 are a lot of technicals errors.
Regards,
Piet Hompus.
Educational Services.
Nieuwegein
The Netherlands.
My review remarks are below this line, I placed them between "(" and ")":
------------------------------------------------------------------------------
Page 1-3.a
(I disagree with the three hour, this is not possible when you want to explain
several things.)
Page 1-3.
To enhance performance, you will use the install utility .....
(Only useful when shared at the same time)
Page 1-4.
The last objective
(Why explaining things about VAXCluster nodes ???, leave that for the
cluster training. We also do not tell about networks.)
Page 1-10.a.
. @sys$update:autogen GETDATE REBOOT FEEDBACK
(Should we advice this, or is it better to check the result before rebooting
the system.)
Page 1-11.
Command procedure !!!
$ .
$ .
$ ON WARNING THEN ....
$ ON ERROR THEN ....
$ ON SEVERE_ERROR THEN .....
$ .
$ .
(This does not work, what do you want ??? , because line 2 overules line 1,
and line 3 overules line 1 and line 2.
So leave out line 1 and 2).
Page 1-12.
Command procedure !!!
$ .
$ Purge /keep=4 system:agen$feedback.report
( is system a logical name ???, or do you mean sys$system: )
( File name is wrong , it should be sys$system:agen$params.report)
Page 1-15.
(Also explain what THRASHING is.)
Page 1-20.
Do not use SYSGEN to modify system parameters arbitrarily.
( Give an example when and why to use SYSGEN,
e.g. modifying dynamical parameters (LGI_xxx_xxx .)
Page 1-21a.
.Each process has its own unique virtual address space.....
(The process specific unique address space is 2 gigabytes, P0 and P1).
. the hardware intercepts these virtual address reference ....
(Mention that the translation is done by the CPU, so many translations means
high CPU activity).
Page 1-21.
PFN - physical page frame numbers
(PFN = PAGE FRAME NUMBER).
Page 1-22a.
The memory management subsystem consists of three components:
(This is not complete , what about e.g. AWSA).
Page 1-23a.
. sets the number of balance set slots in the system page table
(RTFM - internals and data structures.
Balsetcnt set the number of entries in the balance set slot array).
. each balance set slot requires 4 bytes of ....
(RTFM .... EACH PTE requires 4 bytes)
(A PHD is a log bigger, think about WSMAX + PROCSECTCNT + VIRTUALPAGECNT)
Page 1-23.
the $SHOW MEMORY output
Process entry slots 43
Balance set slots 43
(Do not edit VMS output, BALSETCOUNT < MAXPROCESSCNT - 2)
(DO not use $ show memory, if you want to show this display use
$SHOW MEMORY/slots/physical)
Page 1-24a.
Because the working set limit represents the amount of physical memory "owned"
by a process....
(The Limit specifies the maximum amount , not the actual amount of page.)
Page 1-24.
A page not in the working set can be in one of four locations:
(This should be "in one of six locations")
( . also mention "in another Working set")
( . in a private page or section file on disk).
(PFN = Page frame number)
Page 1-25.
Paging is not only to and from the page file on disk,
also mention the page movement in memory.
Page 1-26a.
Whenever an image references a page of its virtual memory that is not currently
resident in physical memory.....
(physical memory -> should be in the working set).
Soft page fault
(Also mention that the page can be retrieved from an other Working set).
Page 1-26.
If the physical page is needed for another purpose:
. VMS first writes out its contents to disk
(This happens not at the moment the page fault occurs but is done at the
time the modified page writer is active)
A page fault for a page that is not in physical memory, (that is , in image
file on disk).....
(Also mention , in pagefile, RTL file or section file)
. controlled by system parameter: PFCDEFAULT
(This parameters defines the maximum size , but if the image section is smaller
less blocks or pages will be transfered).
- Free list if page has not been modified since image ....
------------
(Free list if it is not possible to modify the page).
- subsequent read from file is a hard page fault
(subsequent read from the page file is a hard fault).
Page 1-28a.
You might want to share some other details of paging .....
(Leave this part out, we will not have enough time to cover this)
. page fault occurs when a program refers to a virtual address not yet mapped
(Hardware cannot translate VPN -> PFN
(This is not clear , mention when the page is not in the working set, because
it can be somewhere else in memory and therefore mapped.
-If this is a global page fault .....
(Explain what this is, a global page fault)
Page 1-28.
. Page fault occurs ...... (VPN -> PFN)
( mention when it is not in the working set).
Page 1-35a.
MPW_HILIMIT Modified page list maximum limit.
(The number of page on the MPL which will activate the swapper, the list,
can be bigger than MPW_HILIMIT).
Page 1-35
MAXJOBS (is the maximum number of jobs not processes, it is still
possible to spawn).
this is the same for MAXACCJOBS and MAXDETACH.
Page 1-38a.
(Also mention on this page what goes to memory if no qualifiers are used
- e.g. the FILE-ID goes to memory).
(Also mention the qualifier /HEADER_RESIDENT)
Page 1-43a.
The second alinea, this text here does not discribe swapping but TRIMMING.
Page 1-43.
- swapping interacts with processes
(is it not better to say swapping interacts with Working sets).
Page 1-44.
- substained queue length greater than 0.2
(this should be "greater than 2.0")
$ monitor disk/item=all
(Monitor disk/item=all/interval=1)
Page 1-45a.
. if the average queue length .... 0.2 ....
(Again if it is greater than 2.0)
Page 1-46.
. to locate heavily used files:
(add the next line here : $ monitor RMS/FILE=)
Page 1-47a.
Do not explain UCB and IRP and CRB and etc. etc.
(This is internals , no time for it.)
Page 1-48.
. File header block
(Why not the next area's - Header area
- Identification area
- Map area
- [ ACL area ] this is optional
- check area)
Page 1-49a.
The VAX SPM chapter shows .....
(Where and what chapter ?)
Page 1.51a.
Add the next line
( . open the file via a C- WCB)
($ show device/files/window --> display 1 or 1C WCB.
Page 1-53.
Why is this in this module, leave this for the cluster training.
Page 1-55c.
Why an macro program as an example, why not a DCL command procedure
that loops).
Page 1-56.
. Processes are scheduled for execution using of two techniques:
( it is always priority driven)
Page 1-58a.
MAXJOBS=n
(not the maximum processes but JOBS).
Page 1-58.
- modify/noaccess=(primary, 18-8, secondary, 9-17)
(this line is confusing with Access alowed from 9 a.m. till 6 p.m.)
(Why not using /ACCESS=....)
Page 1-59.
- Also make sure they use locality to reduce page faulting
(Explain locality or do not mention it)
Regards.
Piet.
|
| 106.9 | Review Comments Implemented | HARDY::MOSTEIKA | Paul, ZKO1-1/D42 DTN 381 (881)-1075 | Tue Dec 17 1991 15:50 | 14 |
| You will find a new version of the Introduction to Performance Management
in the review area on SUPER.
SUPER>dire ES$REVIEW:[SYSNET_III]SYSNETIII_CHAP10.PS;2
SYSNETIII_CHAP10.PS;2
4591 17-DEC-1991 15:39:14.60
I've implemented almost all of the comments.
Thanks for all the review comments (even the RTFM, I'm still looking for this
particular manual...).
Paul M.
|
| 106.10 | | SUPER::MATTHEWS | | Wed Feb 26 1992 15:54 | 13 |
| Ooh, I found one!
Page 12-75, step e:
SHOW PROCESS/ACCOUNTING pid
should be either of:
SHOW PROCESS/ACCOUNTING/ID=pid
SHOW PROCESS/ACCOUNTING name
|
| 106.11 | A tough review on Performance Intro | SOAEDS::TRAYSER | Seniority means a bigger shovel! | Mon Mar 02 1992 14:48 | 152 |
| Chapter 10 --
This is the Performance Management Intro chapter. Before I critique it, I
need to say that I teach VMS performance as often as I can, it is by far my
favorite course. As such, I'm quite comfortable with the topics and the
material in this chapter (much 'lifted' from the VMS Performance Mgt course).
My main concern is how fragmented it is, jumping around within the topics
(such as memory management) without any obvious direction, duplicating
information on pages only 2 or 3 pages apart. Second major concern is, that
much of the technical information has either been discussed in previous
courses, or it is too in-depth to teach in a 2-3 hour window! Personally I'd
like to see this module reduced significantly, or even the entire topic of
Performance moved out of SN 2 (and 1?) and using SN 3 as the introduction. The
topic is complex enough and in-depth enough to be kept primarily in the
specialty course.
Overall the material is good, has a good approach (Concepts, Memory, CPU, IO,
etc.) and is relatively error free. These are not my favorite pages out of the
Performance course, nor are these the pages that will give the student the best
*task* oriented approach to performance. If we stay with these basic pages,
then I think rearranging them may give a smoother delivery.
10-6 - 10-8 --
Leave this stuff out. The students are 'task' oriented. This info is
best left for the VMS Perf Mgt course where "philosophy and architecture"
have a place, let's get to the real stuff on page 10-9.
10-10, 1a --
If all you want is a current FEEDBACK data file collected, there are
two easier ways to get this: 1) $ @SYS$UPDATE:AUTOGEN SAVPARAMS or
2) $ Run SYS$SYSTEM:AGEN$FEEDBACK. First, AUTOGEN, when provided with
only a P1 (start phase) will begin and end at that phase. And, using
FEEDBACK on P3 (Execution type), buys you nothing if your end phase is
SAVPARAMS. Second, all that is really getting done in that phase is the
AGEN$FEEDBACK.EXE file is being run.
10-10a, 3rd sentence --
"...and if you're in agreement CORRECT the tuning..." probably needs to
be "...COMPLETE the..."
10-11 - 10-13 --
Cute, but overkill. Put the basics of this on 1 page and put the whole
procedure in an Appendix. There is nothing to TEACH here for 3 pages that
was not explained on the previous page. This is *NOT* a DCL programming
course, and that is all these pages can really be used for after doing a
good job on 10-10.
10-14 - 10-15 --
We are being a bit high-minded here. Assuming we keep the VMS System
Performance Mgt course (in some fashion), we should expand this info
there. These pages should basically cover the definition of Capacity,
etc. then move on to the specifics of VMS.
10-16 - 10-19 --
Not really useful here. This Introduction chapter is turning out to be
a Concepts and Architecture course. Let's eliminate these and leave them
in the performance course.
10-21, 2nd bullet, 3rd dash --
Add: SYSGEN> WRITE ACTIVE ... or something similar. The example leads
us to believe that the SET commands makes the change. Of course I will
usually suggest running AUTOGEN, but they proably need to know how SYSGEN
works.
10-23, 1st line --
We defined a "page" on 10-22 (and probably in SN1 & SN2), so we don't need
to do it here.
10-24 --
We need to introduce the relationship between MAXPROCESSCNT and BALSETCNT
at this point. It is important that these two be discussed and defined
together!
10-29, 1st bullet --
Change "$ R image_name". 1) get rid of "R" and replace with "RUN", 2) and
pick a sample program instead of "image_name", like maybe AUTHORIZE or
SYSMAN or some other program they are familiar with that uses a RUN
command.
10-31 --
Wait!! We haven't talked about working sets yet! How about putting page
10-37 in front of this page. As a matter of fact, this whole section
would probably teach smoother with an order like:
22, 23, 24, 25, 26, 27, 28, 29, 30, 37, 33, 34, 31, 35, 32.
page 36 should be cut up and dispersed to the appropriate pages (i.e.
put PAGFILCNT on 27, SWPFILCNT on 44, etc.).
10-33 & 10-37 --
These two pages are rather redundant. Combine and thin out duplicates.
10-38 --
Automatic Working Set decrementing is NOT something these people should
fool with. Its a crock as far as tuning (that's why it has been turn off
almost ever since it was created), and besides, the Ticker and Troller
eliminate the need for this feature with great success! Save the
'decrementing' topic for Performance Mgt. Also, we really don't need
this picture, as it is very easy to draw and references AWSA-decrementing,
so kill the page.
10-34, 3rd & 4th bullets --
Why is INTERACTIVE in bold? Batch, Network, etc. all use WSQUOTA and
WSDEFAULT.
6th bullet --
Note: Authorize can adjust WS numbers up and down, but SET WORK can only
set them down and back up to the AUTHORIZE limits, but never over.
10-44 --
Should be either in front of or in back of the WS discussion. I'd suggest
putting it either side of page 38. With 5.5 swapping has become more
reasonable, so it is actually a more valid memory manager that AWSA (WS
decrementing).
10-45 --
Fine, but the follow-on page, 10-49, is too far away. Move 49 up to
follow 45.
10-48, 3rd bullet --
"Copy the file, then define the logical" only works for some, not all,
for example: JBCSYSQUE.DAT logical is referenced at the START/QUE/MAN
command. These logicals should really be put in SYLOGICALS.COM.
10-55 - 10-67 --
CPU pages don't lead you to the next topic of discussion, they kinda fall
over each other. Try this order for a smoother teach:
57, 58, 59, 60, 62, 63, 64, 65, 66, 67, 55, 56, 61
10-55 --
Why "balance" before we look?
10-56, 1st bullet --
What is the significance of "/BASE_PRIORITY=n" having 1 "n" and all the
other qualifiers having 2 "n"'s?
10-68 --
Good Grief!! A six page summary?! I'd used it if it was on 1 page, but
I'll never use this one. Don't "review" the chapter, just summarize it.
Sorry for the tough words, but this module needs a fair overhaul to fit
properly in SN3 and NOT get so deep into performance that you can't cover
it properly in 2-3 hours. Let's keep it introductory, task-oriented,
basic commands -- something they can DO when they get home without the
architecture and the concepts overwhelming the practical. It's tough to
teach Performance without these, BUT, to do it right you need more time
than what we have. Keep it simple, and short.
$
|
| 106.12 | put it in earlier | TEACH::RINE | | Thu Mar 05 1992 14:27 | 9 |
| Module 10 should really occur earlier in the book. There are a
lot of complex topics that really should be covered before Friday.
I usually like to cover all of the topics as the last lecture of
the day, and leave all of the drawings on the board. This gives
the students the rest of that afternoon to think of anything that
might bother them about the concepts, and gives them an opportunity
to ask any questions first thing the next morning. I will probably
continue to do this lecture thursday afternoon as I did during the
first teach.
|
| 106.13 | Performance, yes earlier. | SUPER::MOSTEIKA | Paul, ZKO1-1/D42, 381 (881)-1075 | Mon Mar 09 1992 09:45 | 14 |
| I agree with you. I wanted it moved up somewhat. Because on Friday everyone is
thinking about going home. I think it's a pretty impotant topic. You see how
much my vote counted. I guess something has to be last!
Performance requires knowledge of various areas of the system. That is why it
was positioned where it was. At least that was the arguement. It's a valid
argument, but there are still other topics that could be last. Still others felt
that it's not that necessary. That's one of the reasons why it fell out of
SYSNET II.
One positive aspect: it could be used as an incentive or motivation technique
to hold interest until the very end of the course.
Paul
|
| 106.14 | | SUPER::MATTHEWS | | Tue Mar 10 1992 13:31 | 5 |
| Is there another "best" chapter to put last? A big reason for finishing
with this one is so you can end the course with a crowd-pleasing topic,
but that's not to say we can't come up with a better order.
Val
|
| 106.15 | other thoughts | SOAEDS::TRAYSER | Seniority means a bigger shovel! | Tue Mar 10 1992 16:51 | 7 |
| Finishing with Performance is OK. But, there is too much stuff, we need
to thin it down and push some of the less practical stuff back to the VMS
Sys Perf Mgt class. I'm going to try and rearrange my lecture next time
I teach it, I'll probably move Perf Mgt to Thursday...I'll let you know
how it goes.
$
|