Conference bulova::decw_jan-89_to_nov-90

    Yes, a MicroVAX II certainly can use 14MB - at least my system
    uses 13MB quite handily.  Get the 14MB - you won't be sorry.
    A 6MB system is adequate, but 14 is much better.
    
    				Steve

    Re .1
    
    But is it twice as good? Bearing in mind that the 14 Meg version
    is twice the price of the 6 Meg version only half as many engineers
    can have one if I go for the 14 Megs.
    
    Given that I will have 8550/6220 in the same cluster as these beasts
    do I need 14 megs on each desk? Remember I have a local disk
    for paging.
    
    Does anybody have emprical figures?
    
    Dave

	Not empirical, but from memory:

	VMS	= 1.5 meg
	DECnet	= about 1 meg ?
	LAVc	= 1+ meg
	DECwin	= 1+ meg
	Total	= about 5 meg

	This workstation (4 disks, 2 ethernets, tape drive) has 6 
	megabytes in use when nearly idle when not in a LAVC:

              System Memory Resources on MYNODE 20-FEB-1989 20:24:10.94
Physical Memory Usage (pages):     Total        Free      In Use    Modified
  Main Memory (13.00Mb)            26624       14136       12265         223
Slot Usage (slots):                Total        Free    Resident     Swapped
  Process Entry Slots                 25           8          17           0
  Balance Set Slots                   22           7          15           0
Fixed-Size Pool Areas (packets):   Total        Free      In Use        Size
  Small Packet (SRP) List            399          81         318          96
  I/O Request Packet (IRP) List      250          31         219         176
  Large Packet (LRP) List             51          12          39        1648
Dynamic Memory Usage (bytes):      Total        Free      In Use     Largest
  Nonpaged Dynamic Memory         428032       32704      395328       28816
  Paged Dynamic Memory            174080       45072      129008       43968
Paging File Usage (pages):                      Free  Reservable       Total
  DISK$3BROS_19668:[SYS0.SYSEXE]PAGEFILE.SYS                                    
                                               28770        3413       29992
Of the physical pages in use, 3324 pages are permanently allocated to VMS.

On my node (part of a LAVc with 10 other members), I run 5 DECterms, VUE,
Cardfiler, DECwindows Mail, Calendar, DECwindows Notes, Fish, Kaleidoscope, and
5 DECterms.  I now have 13 Meg and things run comfortably.  I used to have 9 Meg
and things were very pinched--I could not have survived without periodically
running C.W. Hobbs's FREEMEM program to cause everything to purge its working
set.  6 Megs would have been out of the question.

If your prospective users are just going to run a few DECterms logged into big
clusters elsewhere, then 6Megs will probably work out just fine.  If they're
really going to USE DECwindows on their 'stations, go with 14 Megs.
Particularly since you're talking about running them in a cluster.

--PSW

    Just something that you might want to consider.  Eventhough you would
    only have 6 Meg of memory locally, you could start most of you clients from
    a remote machine (a system on the cluster with memory resources) and
    just display them on the workstation.  You would still want to run File
    View locally and maybe a couple of DECterms.  After that, everything
    should be performed remotely.  If you need more DECterms and if this is
    an DEC internal environment, you could use CHILD for more DECterms.
    
    As far as running everything locally,  I've spoken with people who
    upgraded their VS2000's from 6 meg to 14 meg and that stated that the
    performance was much better.  Much less paging on the system.
    However, is it work the price of a 12 Meg board? Hard to say.
    
    David
    

    When running the OOTBs, the 'break-over' point seems to be about
    10mb - which means that having more than 10mb buys little. (this is
    medium usage, not super heavy usage)
    
    Systems under 8mb seem 'really really' tight. (read: slow)
    
    I run successfully with a 6mb 2000 because I run everything but server,
    session manager, window manager and DECterms remotely.  That assumes
    that the remote machine has plenty of memory - if it doesn't forget
    running remotely, you'll just kill it.  Also, PAINT doesn't run
    remotely very well, the brush has trouble keeping up with the mouse.
    
    The primary resource consumption of DECwindows is memory, not CPU or
    even I/O. The primary tuning advice is "Buy More Memory!" Seriously.
    
    Bottom line = 
    
    	IF .NOT. 3000-SERIES THEN
    	BEGIN
    		IF .LT. 8MB THEN
    			RUN OOTBS REMOTELY
    		ELSE
    			RUN OOTBS LOCALLY
		ENDIF
        END
    
    	{YOUR MILAGE WILL VARY}  (being in cluster hurts, of course)
    
    
    Comparing the 6mb vs 14mb 2000s I've seen - it's worth it!
    
    	Ken Henderson

    I run only DECW$BANNER and one DECterm on my 6mb VS2000, all the
    rest (5 DECterms, EPICwriter, calander, mail) all run on a remote
    node (VAXserver 3600 Series, CVAX type thing with 16mb). This leaves
    me with ~1000 pages of memory free and performance is quite resonable.
    
    
    Stephen.

I run a couple of DECterms locally, but FILEVUE and all that it starts on our 
8530. Performance is ok, but sometimes the ethernet gets bogged down (like
right now for example when my typing is several characters ahead of the words
showing on the screen). Running more clients directly on the 6MB VS2000 soon
bogs it down. Our group has now ordered the 14MB upgrades for all our VS2000's.

Tim

I ran for several months with a 6 meg vs2000 in a cluster with a 8550 boot node
and remote pagefile.  I ran almost all applications remotely, generally starting
up only a single decterm so that I could submit a batch job that ran VUE.

Then my vs2000 was upgraded to 14 meg.  Any time that I tried to run just some
of the applications locally, I'm disappointed with the performance and go back
to running everything on the boot node.  Recently the 8550 was replaced with
a 6220 and I still find the remote applications better when remote.

A comment about the `pauses' noted in .8; I also noticed this behavior until
I setup a batch queue that ran at priority 4 instead of 3.  That might be the
source of your delay.

I have had good experience using VMS Ultrix Connection aka TCP instead of DECnet.
It seems to be less sensitive to the delays caused by quorum wait.

I have looked at the traffic on the wire generated by DECwindows and it looks
to be similar to terminal traffic in character, small packets with an arrival
rate on the order of 3-10 per second.  This is probably not much heavier than
DECserver100s used in an engineering timesharing environment, ie., the LAT
connections do do much multiplexing in practice (because the number of lines
is so small).  It really doesn't make much difference if a key stroke is carried
in a 60+ byte packet as one byte versus being carried in an 80 byte packet as
a 40 byte event (I have no idea what is in each packet; I'm just guessing).

Now, one might argue that if the application were run locally then the boot
node would have to deal with the network interrupts, the context switching of
the application and/or terminal emulator, etc.  Hwoever, it will have to deal
with the disk I/O serving caused by the the application running on the
workstation.  My argument is that I expect that a workstation will give me
performance equal to or better than what I got as a timesharing user as well
as give me additional functionality.  Since I got very good service on a 6vup
cpu with fast disk, there is no way that I'll find a vs2000 comparable no
matter what the disk.  The situation might be different if your 8530+6220 is
memory and CPU bound to the point where you are only able to get 1-2vup and
10-20 i/o per second, in which case, you I/O from the workstation will take
priority over the local I/O on the boot node and will probably result in no loss.

Again, my subjective opinion is that I gained no performance when going
to 14 meg.

    This may even be shaping up into a consensus.  I run everything but
    fish, etc. remotely on an 8800.  Performance is just fine.  Running
    VUE, I mean FileView, remotely works just fine for me, though I think
    the CPU priority comments are worth noting.  This is all on a 6M VS2000.
    I just saw 2250 pages free, though I have 5 DECterms, VUE & clock.
    (Start up a couple more apps and watch that memory vanish!)
    
    It would seem your primary concern would be where your users will be
    running their jobs.  Which in turn probably depends on what kind of
    boot server you've got.  I like the notion of a behemoth 8800 for the
    hard work, and VS display for the presentation.
    
      John

	I had a 6MB VS2000 and it was relatively useless as a development
	machine.  With a recent 12MB upgrade it is reasonable.  My 
	organization received 13 6MB VS2000s and all 13 are getting the 12MB 
	boost since we've seen the difference in performance.  Believe me
	if you want to any local work (i.e., compiling etc...) the 12MB
	upgrade is worth it.

	Jim

    	Well, I'm reporting from Hardware Heaven as far as the other
    folks in this topic...
    
    	WAYLAY is a 16 meg standalone VSII/GPX. Dual RD54 and about
    13 million blocks of DFS disks.  Both DECwindows and VWS installed
    on the system disk.  Lots of LP's.
    
    	I run 6 DECterms, Fileview, Session Manager, Banner, Calendar,
    two Emacs (subprocesses) locally.  Sometimes worm, or kaleid, or
    fish.
    
    	I run Fileview & Bookreader from the 8700 in our cluster.
    
    	I've had it down to under 1000 free pages!  I average 30-40
    processes.
    
    	All our workstations (GPXs) in the cluster have 13 meg.  When they
    had only 9, they ran like dogs.  Our one 6 meg VS2000 [RD54] is painful
    to work on.  All the workstations have local disks for paging and
    swapping, as well as user disks.  The user disks are not served
    to the cluster.
    
    						--Doug
    
    
    

    I thank everyone for their input.
    
    My conclusion is that 6 Megs will hack it. But there should be
    some powerful systems backending the cluster. We don't really want
    to do a lot of development on the VS2000s. I believe that a microvax
    2 is CPU far too slow for that. Better to kick of compiles on the
    big machines.
    
    It appears people have successfully run 6 megs. Yes it looks like
    some tuning is needed.
    
    Dave

RE .7

How do I run DECterm remotely?  I get the following message.

(Please don't tell me "Do as it says!".)

$ set display/create/node=JIT631
$ RUN SYS$SYSTEM:DECW$TERMINAL

DECterm version V1.0-1 now at your service...

To create DECterms, use the Session Manager or call the
DwtDECtermPort() routine from another process.

Try the DECTERM conference.  Press KP7, etc.

It's at: HANNAH::DECW$DISK:[PUBLIC]DECTERM

Be forewarned, however, that there are a few entries already that state 
EXPLICITLY how to do this.  If you ask the question without doing the research
it's unlikely that you'll get a friendly response.

	- Dave

Re: .15

As stated in reply .15, you should look in the DECterm conference.  Look
at note 289.1.

				-- Bob