Conference koolit::vms_curriculum

    Chapter 2 on Memory Management has been revised and placed in the
    review area.  Please place comments in this note.
    
    ES$REVIEW:[VMS_PERFORMANCE_V55]VMS_PERF_2_MEMORY.PS
    
    
    Note a new format for LL has been used.  I may or may not have this
    format totally correct.  You will notice each chapter has a chapter
    terms section and a two part summary.    Also notice that the chapter
    is information mapped.
    
    On the instructor pages I have given the description for each parameter
    from the sysgen manual. 
    
    Emmalee

    I have placed a new copy of Chapter 2 in ES$REVIEW.
    
    The first copy did not have  "INSTRUCTOR NOTES" at the top of each
    instructor page making it hard to review.  
    
    Also I have received comments from two people and I updated the chapter
    with their comments.  Please send yours  by May 22 for the pilot
    edition of the course. 
    
    I have received good and helpful comments.  I need your comments too.

Here are my comments/thoughts on the original Memory Chapter.
    
    (ie: *not* the version mentioned in .2 . I copied mine on 12th May and have 
    just finished typing the lot out.  Why didn't the 2 people who have sent
    comments post them in this note? That may help other reviewers (like
    me!).
    
I haven't bothered with insignificant typos on the instructor pages, but have
pointed out those I spotted on the student pages.


I like the new order, it flows better than the old one.

Page 3:		Introduction:  why mention "the speed at which the CPU can
		access the data"?  It's not mentioned anywhere else in the 
		module and can't be changed.

Page 4:		Remove the section headed "TIME" from the student guide.

Page 5:		Chapter terms: 
		
		12 lines down:  User Identification Code
		needs caps on "I" and "C".

		installed images:  "A program" should be "An image"

		page fault: "." missing at end of second line.

		paging file: "." missing at end of second line.

		process header: "stsructure" should be "structure"
				"(PCB)" should be "(HPCB)" as (PCB) by 
				convention means the Software PCB.

		process header slots: "addressspace" should be "address space"

	[Anything wrong with the pagination at the end of this page? It
	overflowed by 4 lines on my printed copy]


Page 9:		line 14: "portions data" should be "portions of data"

Page 24:	add "idle" or "busy" to the batch queue display to pretend 
		it's a 5.5 output.

Page 26:	top line: remove "," from after "limit", and place it after
		"additional page". [sentence then scans better] 

		last line: replace "in physical memory." with "available for
		the original process's use."
		[I assume you mean that a page has been removed from the free
		page list to be added to some other process's working set. It
		will still be in physical memory.]
               
Page 27:	6th bullet point: Add at the end "(Soft page fault)".
		[similar to page 26 and the free page list fault]

Page 29:	3rd bullet point:  "WSDEAULT" should be "WSDEFAULT".

Page 36:	Can you obtain some "real" figures for the effect of TB
		checking?  Also which processors is it implemented on? The
		instructor notes page says "It is not used on all processors".

		The warning that "too large" or "too small" can cause
		performance degradation when TBSKIPWSL could be anything
		between 8 and 512 begs for more info on setting this parameter.
		Any guidance here would be beneficial. (Well, it'd certainly
		help me!)

Page 39:	5th bullet point, first "-" :  Writing out the modified page
		list to disk is subject to two conditions. It may not,
		therefore, take place and SWAPPER may go straight to looking at 
		trimming working sets. [See next page for details].

Page 40:	2nd bullet point: should be:  "If pages exceed MPW_THRESH *AND*
		the (FREEGOAL - current freepage count) deficit can be
		recovered, write the modified pages to the paging file and
		place the pages on the free page list."
       
Page 44:	line 12: add at the end of the line after WSQUOTA: 
		", had a recent PIXSCAN boost ".

Page  45:	6th bullet point: "working set" should be "working set limit".
		7th bullet point: "working set" should be "working set limit".

		Reclaiming Memory from Periodically Waking Processes:
		Add an extra bullet point:
		"Activated when the size of the free page list drops below
		twice the value of FREEGOAL"

Page 46:	IRPSIZE is not a SYSGEN parameter.  In two places in this 
		diagram, "IRPSIZE" should be replaced with the value "176".
		[This is a hard coded value]

Page 47:	Update the $SHOW MEM/POOL/FULL display to VMS 5.5.
		[The one on page 15 is VMS 5.5]
		ie: SRPSIZE is 128
		IRP Lower bound is 129
		LRP Packet size is LRPSIZE + 352, making it 1856.

Page 49: 	Can you update the date from 1988?

Page 50:	ditto 1989?

Page 51:	ditto 1988?

Page 52:	ditto 1988? [The format of $SHOW PROCESS/ACCOUNTING has changed]

Page 54:	3rd line: shouldn't "server-client" be "client-server" ?

Page 57:	2nd bullet point: "open files" should be "open image files".
		6 lines from bottom of page: remove "virtual".
		[All processes sharing an image have a virtual copy of the
		program. Installing the image /SHARE stops them having 
		a separate physical copy]

Page 58:	7th bullet point: 
		Add "Header resident implies Permanently open."

Page 60:	Empty except for the heading "Memory Savings from Shared Code".
		Diagram missing?

Page 62:	3rd line: ">INSTALL" should be "INSTALL>".
		7th/8th lines: Multiple successive calls of the same image is
		ineffecient and bad practice [Especially on a performance
		course! And the example is of a startup file !!! ;-) ]. 
		
		Replace with
		
		$INSTALL
		CREATE etc
		CREATE etc
		CREATE etc
		$

Page 63:	Top line of numbers in circles: "4" should be where "5" is
		Last line of numbers in circles: "6" should be "5" and
						 "7" should be "6".


That's it!


Keep up the good work!


Geoff

Perf Chapter 1 (memory module) -- review

Basic comments...

  VERY NICE JOB with the Instructor pages.  These were very clear and
 useful in most cases.  Great help for a successful prep or for answering
 the oddity questions that always come up in class.  

  I think we jump around between working sets and paging/page faults a
 bit much.  There is a tendency to describe the details that are needed
 before we get to the pages with the details because of the interceding topics.
 Overall rather teachable.

 1-4 --
    The "TIME" stuff is unclear.

 1-5 --
    The "glossary" is great!  But so we want it at the front or the back of
    the chapter?

 1-7 --
    Remember that are outside Consultants that teach the course but don't
    have access to the Instructor's book.  The may not know what is meant
    by "Internal" and "External" and "CPU" memories.  Maybe adding "Registers"
    "Main Memory" and "Disk" behind those definitions would help.

 1-8, 8th bullet --
    High-speed cache is sometimes included (see uVAX II).  To say that it is
    included assumes we know which systems and how much is there.  

      9th bullet --
    Implied here are two items -- that we can 'influence' it and that we can
    control it.  Both are false from the system managers point of view.  In
    the best case we have a system with a large I&D space and s single 
    application.  

      10th bullet -- 
    Fine.  How do I see the HW cache hit rate?
 
  1-9, Program Variability --
    How much of a programmer's code can be in cache?  The space is relatively
    small.  Does the effectiveness vary with realtime vs timeshare?  Is there
    an optimal Working set size?  To be honest, the programmer needs to write
    the programs assuming there is NO cache on the system -- careful use of
    Working set and virtual address requirements.  Are we getting down a
    "Programmers Rat Hole" that is not really pertinent to System Managers?
    Most of my students are non-programmers or very casual programmers, as
    such there is nothing of much use in this discussion so far.

 1-10a --
    "Why list the 6200?  Do all the 6000s use two levels of cache?  If so, 
    drop 6200 and list 6000.  Also, let's put the 9000 (and 4000s if they 
    have it) on the list.

    The Modified List and the Free List are referred to as caches in VMS.
    We need to clarify that there are caches in Main Memory, not just the
    separate high-speed memories.

    The TB cache is most effective as a temporal cache -- recently accessed
    pages.

 1-9a --
    In referring to cache misses in the first paragraph we are assuming that
    1) we have a system that implements caching in the HW and 2) that there is
    something that we can either watch or manipulate.  How do we monitor the 
    effectiveness or even detect the presence of this cache?  For Row_Major
    this is best demonstrated on the Freelist and Modified List caches -- not
    HW cache. 

 1-10 -- 
    This diagram would more effectively introduce the topic of caching.  Maybe
    place this in front of the Cache discussion.

 1-12, 2nd sentence --
    "...itself a COLLECTION of programS and..."

 1-13a --
    Any particular reason for listing just Freelim and Freegoal for the Free
    List parameters and High and Low limits for the Modified list??

 1-14a --
    What are the "(G,M)" entries??  Do we need them?

    Listing the defaults for MAXPROCESSCNT and BALSETCNT is useless since on
    a standard installation AUTOGEN gets run automatically and sets them to
    different values.

 1-14 --
    Spacing problems with the Date and Modified pages value -- should be 
    further to the right.

 1-15 --
    Many instructors and customers are not aware that VMS has changed its
    name again (formerly VAX/VMS) to OpenVMS (or what ever they decided
    is the final, official, not-going-to-change-again name for VMS.  We need
    to specify this in the Introductory pages of the book.  In the same
    line of thinking, OpenVMS can also describe VMS for Alpha which does
    not use the same Physical/Virtual translations.  So we need to specify 
    that this course is currently targeted for VMS for VAX.

 1-16a --
    Virtual addressing is limited to 32 bit addressing, although physical
    is increasing to 34 or 36 (was it in Blade?).

    Last sentence of the paragraph: "Much larger that the physical memory
    of *any* computer and much larger than *any* program needs."  These are
    both incorrect.  This might have been true 3-5 years ago, but look at
    the specs for Alpha systems.  Let's 'water down' the "any" words.

 1-16, 4th bullet --
    Pages are numbered normally in Hex.  So put "(800,000 hex)" behind the
    8,388,608
 
 1-17a --
    formatting -- trailing open-paren on second sentence.

 1-19a, 2nd sentence from end --
    "...can not NORMALLY access..." -- If you have sufficient privs you can
    do most anything.
    
    By the way, VMS engineering doesn't usually refer to the Reserved Region
    of address space as S1 -- this is 'slang'.

 1-20a, Title --
    We aren't really talking of Paging as much as we are of Working Sets.  Can
    we drop the PAGING title?
 
    We might want to note that Paging will be more completely discussed on 
    following pages -- don't go into details of paging until then.

 1-20, Title --
    ditto with title of page.

       6th bullet --
    "...the limit."  Which limit?? It is unclear - specify Working Set limit.

 1-21a, spelling --
    quantum is misspelled without the 'n' on the 3rd paragraph from the bottom.

 1-21, 7th bullet, 2nd dash --
    At this time we don't need to be telling students that we can get pages 
    from the loan regions of other processes.  This is a Working set concepts
    discussion not a reclamation discussion.

 1-22a --
    We may want to state that the PQL_MWSxxx parameters limit the low settings
    of the working sets (overriding SYSUAF).

 1-22, 1st line --
    "A process running interactively from a login process" -- strange words,
    we need to be clearer on the student pages than this.

    Also, the bullets partly obscure other numbers.

 1-23a --
    "This need not be emphasized, EXCEPT FOR THE FACT THAT THE VALUES OVERRIDE
     those set by AUTHORIZE/SYSUAF."

 1-24a, "demand-zero" --
    On page 1-21a we called it "demand" not "demand-zero".  Are we talking 
    different terms, should we make a clarification?

 1-25a, 3rd paragraph --
    "...keeping track of which a page was least used." -- the "a" seems out of 
    place.

    "The VAX does not have a use bit, ALTHOUGH IT DOES USE TBSKIPWLS ON SOME
    SYSTEMS."

 1-25, title and FPL --
    Seems "Adding Page to a Full Working Set" would be better than "Removing
    a Page from the Working Set".

    "The free page list is a list of pages that have recently been replaced 
    from the working sets of multiple process."  This is only partly correct.
    On a freshly booted system, all pages 'start life' on this list, never 
    having been in a working set yet.

 1-26, linker options --
    maybe a reference on the instructors page as to where to send the students
    for more details.

 1-27a, typo and "large" --
    "...consists of OTHE free and modified list."  Should this be "THE"?

    1st bullet says keep a large FPL.  But, define large.  Since of an average
    working set?  1/2 of physical memory.  I'd guess what you are trying to 
    say is depend on caching on the FPL, which is OK, but still intensive on
    the VAX systems.

 1-28a, reruns --
    We've already seem similar on pages 1-20 and 1-21

 1-30a --
    2nd bullet should read "Pages are read in increments called clusters in a 
    size up to PFCDEFAULT"

 1-30, "Two system parameters" --
    We left off the PQL_xWSxxx parameters.  

 1-32a --
    1st sentence, "...only way for VOLUNTARY decrementing..." because the 
    swapper could out-swap the process body, that would decrement the memory.

    2nd sentence, PFRAL is missing the "T"

    3rd paragraph, (Nit) "...processes whose page..." - a little personification
    here seems strange.  How about "...process which have a page..."?

    4th paragraph, HUH!??  Never heard of this, nor has my internals instructor
    heard of this.  Is it new with V5.5?  Where is this documented??  This is
    *VERY IMPORTANT* since I am totally unaware of this feature--please advise!

    5th paragraph starts with "WSDEC(D,M)", do we need the '(D,M)'?

    Last paragraph, "V%" should be "V5". 

 1-32a (2nd page), LABELLED has too many "L's"

 1-34a, 2nd paragraph --
    So, which systems implement this and which ones don't?  Can we get a list?
    Hallyburton suggested asking in the EAGLE::VAX conference.

 1-35, FPL diagram --
    This order is no longer 'normal' of the new mem-mgt routines (MMG_CTLPAGES)
    features are used.  FREEGOAL is much higher and FREELIM, GROWLIM and 
    BORROWLIM are all the same down under 100 (63 on my system).

 1-36a, 2nd bullet --
    GREAT STUFF!!!!  JUST WHAT I LIKE TO SEE -- a reference where I can go 
    look and ask!  Keep it up!

 1-36, "MPW_WRTCLUSTER"
    Can we fix the wrapping?  Moving the text columns over by 1 or 2 columns
    should do it!  I've asked for this 2 other times!!!

 1-37a, FREELIM --
    "...minimum number of pages that must be on the free page list." is not
    correct.  It is the minimum number of pages that the system allows on
    the FPL before beginning memory recover procedures.

 1-37, 4th bullet --
    Topic is Swapper Trimming and the bullet says "Initiates swapper 
    trimming..." sounds weird.

    5th bullet, why is "Obtain" beginning with a capital "O"?

 1-38a, last two sentences --
    "It is not just a change in limit.  The limit is not changed."  Huh?
    What limit.  Is it "not just a change" or is it "not changed"?  I
    don't understand!

 1-38, 2nd bullet --
    It is unclear that this is only done if 1) FREELIM is encountered, and
    2) if the difference between the current number of pages on the FPL and
    FREEGOAL can be satisfied by the pages above MPW_LOLIMIT.

 1-39a, 2nd sentence --
    "On the theory is that..."  wording is weird.

    3rd sentence --
    It says "*Two* criteria..." and you list 3 bullets.

 1-40a, first bullet --
    I thought a Real-Time process couldn't get an increment above WSQUOTA
    anyways?

    Last lines -- dormant was already defined on page 1-39a about 1/2 way
    down the page.
 
 1-41a, spelling error --
    first paragraph has "systemwide" as one word...try a hyphen or two words.

 1-41 --
    first line -- There are now *three* strategies...see #3 on page 1-42!!!

    4th dash -- here it is again.  This *must* be new stuff, please give me
    a pointer to the discussion!

    5th bullet -- states "Digital layered products" -- I don't care what 
    vendor wrote it, commonly the 3rd party people pick just as good SW numbers
    for their stuff as we do.  Drop "Digital"

    Last bullet, last 1/2 of statement -- this is not true in practice.  I've
    done a fair bit of tuning and a process that spends a lot of time playing
    with MPL cache *will* degrade performance, sometime greatly.  If it plays
    mostly with FPL then the damage is minimal.

 1-42a, 1st line --
    Are we "VMS" or "OpenVMS"?  Personally I'd like us to say in the intro that
    both terms may be used, but references to VMS actually mean OpenVMS.  (Gad,
    I hate that name!)

 1-42 --
    3rd bullet -- what defined a deficit?

    7th bullet -- "original" should probably be "previous", as original can
    be easily confused with original 'process just created' values.
    
    9th bullet -- How many times does it trim?  Down to what threshold?

    overall this page could be a bit cleaner, for example there are no 
    references to FREEGOAL (or FREEGOAL*2) values used by the routine!  See
    the V5.5 seminar for some good pages to copy from.

 1-43, IRPSIZE --
    IRPSIZE is misleading, how about "Size-of-IRPs" or 176.

 1-44a, 2nd paragraph --
    these regions do not shrink at all!  (until reboot!)

 1-44, Version! --
    get a current (V5.5) display and make sure there is some overdrafting!

 1-46a, MPW_WRTCLUSTER --
    default is 120, not 96.

 1-46 is split on two pages.
    
 1-50a, last line --
    Isn't this the same formula that MMG_CTLFLAGS sets things to??

 1-50 --
    1st line should be Client-Server, not Server-Client

    3rd bullet should say "...or on separate SYSTEMS." since a VT1000 (an X
    terminal) is basically a smart terminal that runs the X-Windows SERVER.
    It is *not* a NODE.
 
    7th bullet, AUTOGEN --  AUTOGEN usually doesn't change this, the various
    files associated with DECWindows startup (i.e. DECW$CHECK_PARAMS) handles
    raising this and other parameters.

 1-52a, 2nd paragraph, last line --
    "...be in the same cluster, BUT SHOULD BE ON THE SAME LAN."  Since DECnet
    point-to-point *will* work, but is INCREDIBLY slow!

 1-53, 2nd paragraph --
    this is really unclear about what you are trying to say with the one-page
    issue.  Can you expand it a bit??

 1-54 is blank!??

 1-55 --
    Can we drop the 1st, 5th and 6th bullets from this page?  They have nothing
    to do with performance and only add to the confusion.  Maybe move them to 
    the instructor's page??

 1-58 --
    Summary ends with Paging and Working Sets, although there is a lot more to
    the chapter that should be 'reviewed'.

    3rd bullet -- "A cache is USUALLY a memory of..."

I finally printed a relatively current copy and have some additional ideas.
I had a chance to lay out the pages on my dining room table like a "story
board", if you are used to the film making process.  I've reconsidered
my initial comment, the flow looks rather good.  Some tough choices were
make, but the WS/paging section looks quite teachable.

Here are some comments after having it all spread out to look at.  These
changes (including a few typos, shuffles and cut-and-paste jobs) I believe 
will make it flow a little smoother and be easier to teach...

1-8 -- 
   Figure 1-1 should follow the first 4 bullets on this page, so to make room,
   move the cache discussion onto its own page immediately following.

1-24 --
   4th bullet should read:
     "the virtual page is ON THE:"

   And don't forget to briefly mention Demand-zero, Global Valid and Write in
   Progress faults...they DO make up part of soft faulting.

1-27 --
   This page is rather useless, not to mention a bit controversial.  I disagree
   with the two bullets on the instructors page explaining it.  Can we just
   drop this page -- it adds very little to the material.

1-31 --
   Missing Figure Number

   Also, this page should FOLLOW the AWSA discussion (page 1-33)

1-35 --
   This page would be more useful following page 1-26

1-45 - 1-48 --
   These pages should be located near the discussions of the topics.  Putting 
   them here requires us to possibly review the previously lectured pages to
   explain what we are doing.  They would reinforce the lecture topics much
   better if they were moved beside the associated lecture pages.

   1-45 -- kill this page

   1-46 -- belongs after talking about FPL and MPL - probably after 1-26

   1-47 -- example 1-5 needs to be near working set diagram - page 1-31
           example 1-6 needs to be near shared memory section - page 1-54
 
   1-48 -- Set/Show Working_set belongs ON page 1-22 with the discussion
           Init/Show Queue needs to be merged with display on page 1-23

$

    I am now working on revising chapter 2 after the pilot.  I expect to
    post the revision Thursday morning.  Because I will have made several
    changes, I will not give this chapter to the editor until Monday.
    
    Please think about the following.  I think that the methods for
    maintaining the free page list should be presented in the order in
    which they "kick in" rather than the order in which they were added to
    VMS.
    
    

  Emmalee, some thoughts...
  
  I'm not sure if that's a good idea.  The 4 major FPL parameters have
  meanings.  With MMG_CTLFLAGS turned off or on it could change the order
  of their discussion.  If we assumed that MMG_CTLFLAGS was shipped "on"
  with all systems, then we encounter 2*FREEGOAL first (Ticker).  Then we
  encoutner FREEGOAL (Troller), and THEN we encounter FREELIM, GROWLIM and
  BORROWLIM all at the same time.  This is NOT a good way to approach it.
  We need to explain the purpose of these values under their NORMAL use.
  The advantage of Ticker/Troller vs. "standard" FPL management.
  
  I know this is a V5.5 course, and that MMG_CTLFLAGS being on might be the
  new default, but many people will turn it off as it forces many MORE free
  pages of memory onto the FPL than many people want on the FPL.  Also,
  most of my customers are still at V5.4 and earlier, so understanding the
  'current versions' is important.  Besides, showing the standard methods
  historically is a very easy teach.  
  
  The explanation of the HISTORY of FPL and MPL is esential to understand
  WHY  SWAPPING was a good way of managing memory in the old days, but not
  in the V4/V5 days.  
  
  Explaining FREEGOAL before FREELIM doesn't make sense, neither does
  mentioning GROWLIM before BORROWLIM.  
  
  [Initially I'm not excited about this change, but will look carefully
  at the material this weekend if you can get it posted before Friday
  morning!  I'm planning on being in ZKO on Monday afternoon, let's spread
  it out on a BIG table and critique the new approach.]
  
  $

    I agree with Buck on this one.
    
    The standard features apply to all Versions of VMS our students have, 
    but the Ticker and Troller are switchable. They should be explained 
    AFTER the FREELIM/FREEGOAL/etc discussions.

    
    
    Make that three votes!!!!!!!!!!!!
    
         GJD

    I have just posted the revised  chapter 2  Memory in es$review
    
    SUPER::$1$DUA6:[ES$REVIEW.VMS_PERFORMANCE_55]VMS_PERF_2_MEMORY.PS
    
    One drawing is being revised as I write this note.
    
    I know I promised change bars, but I get carried away and forget.
    
    I will not send this chapter to the editors until the middle of next
    week so you still have a shot at it.
    
    I have not had time to revise the DECwindows portion.  If I have time I
    will work on it next week.  It appears most likely that this will have
    to wait until the next revision.
    
    Now working on chapter 3.

Chapter 2 has been revised again.  I reposted it in ES$REVIEW last night. 
Please look at it again and make comments today or tomorrow morning at the very
latest.

We now have a discussion the FPL parameters first followed by a section on
Maintaining the Size of the FPL.  Methods are discussed in the order in which
they "kick in" and I think it works very well.  Ticker and Troller names are
not used.  I followed the terminology in the new CUIP documentation.

All chapters have to go to editing on Thursday afternoon.  Firm deadline.

Thanks for all your comments and help.

Emmalee

  Tough module, but I think it's going to work out.  There is a significant
  rewrite of this module.  Plan on re-prepping.  I think it will work fine
  for the 'follow the book' instructors, but people like me will probably
  do what we always have done...use it as a guide and wander off and do our
  favorite dog-and-pony show for memory.  Nothing wrong with this, and I
  think the module will lend itself to this a lot better than the old
  module.
  
  Any comments on Memory Mgt for DECwindows would be useful...post them
  here!
  
  $

Chapter 2 review of final material...

2-7  -- The cache topic needs a bit more meat (some hardware timings, good 
        cache hit rates, more details on the TYPES of caches, etc.) but it 
        is better than the pilot material.

2-9  -- I still think the diagram is in the wrong place.  It belongs near
        "memory types" on 2-7 or just move it to the instructors page and
        let us draw it if needed.  It only shows 1 type of cache, so it really
        has a limited use at this point.

2-12 -- I never liked the 'box' diagram, but now I miss it.  I do hope it is on 
        the instructor's page (haven't printed the current copy, so I don't
        know for sure).

2-15 -- 4th bullet.  So what!  Let's move it to the instructor's page.

2-20 -- "Initial Workins Set Size" belongs on 2-19 after Working Set and
        before Page Faults.

  $