Date: 01-18-91 (09:59)              Number: 3247 of 3248
  To: ALL                           Refer#: NONE
From: PHIL GORDEMER                   Read: (N/A)
Subj: TUNING YOUR NETWORK           Status: PUBLIC MESSAGE
Conf: LANTASTI (5)               Read Type: GENERAL

   This small note is to help answer some questions that constantly get
asked on the best settings for use with Lantastic 3.02.  Please note I
DO NOT work for Artisoft, but do a lot of installations and spend a
significant amount of time handling Network Tech Support.

   Many people have problems getting good performance when running NOS
3.02. Let me clarify here what is happening.  In the previous releases,
2.57, system response was very good on small systems (4 stations or
less, which is what many PCBoard systems are) but speed degradation was
significant as you added more stations, to the point that when you got
to about 8-10 performance was totally unacceptable. When 3.02 was
developed, the underlying strategy was to allow it constant upgrade
paths and consistant throughput as the system got bigger.  This is done
by the added control given through TASKS and Network buffers sizes.
Correct use of the 2 parameters allows Lantastic to operate faster over
a variety of system sizes and loads, BUT because of the flexibility
build into the commands, incorrect settings can severly reduce
performance on small systems.  In the old 2.57 version, the network
tasks and buffers did not work correctly so not much was being changed.

   Below is some recommended rules of thumb to use to get the best
performance from Lantastic with the least amount of work.  These are
general statments and some refining might have to be done on each
system.
---------------------------------------------------------------------
NETWORK COMMANDS in AUTOEXEC.BAT
   General Notes here, Share should be loaded before REDIR.  On 386
   machines, DO NOT LOAD SHARE HIGH.  Major performance degradation is
   seen when many files have to be open. Size and buffer on the REDIR
   command line are important.  You might want to experiment with the
   size a bit.  2048-4096 seem to work best.

   ae2 irq=??
   ailanbios
   share
   redir <machinename> size=3072 buffers=3
---------------------------------------------------------------------
NET_MGR
   Set tasks to (# of nodes / 3) + 1.  DO NOT MAKE THIS # to high, this
      is the best way to kill system performance.  If in doubt, lower it

   Network Buffer - 12K - 16K.  Again, more is not necessarily better.
      Bare minimum is 8K
---------------------------------------------------------------------
DISK CACHE
   Lancache provided should only be used if the server is NOT a station.
      This is because it supports delayed writes and that will destroy
      uploads/ downloads on that machine.  When using Lancache, EXTended
      memory is perferred.  Please note that Lancache does NOT cache the
      directory structure.

   Super PC-Kwik is the all around best.  Use EXPanded memory with the
      following setting.
          superpck /a+ /q+ /h- /d-  (h- and d- ARE MANDATORY)

   VCACHE 5.0 also works well.  Be sure to disable timed writes.
         cache-em /t=0
---------------------------------------------------------------------
DEDICATED SERVER
   If you have a machine that is a dedicated server (IE, you do not run
      anything on it), then you can give a MAJOR performance boost by
      reducing the server time splice to the local application.
      (EXample: On my server I run a local copy of PCBoard, but only
      to start the event. Since it just sits there and ITS performance
      doesn't matter, I reduce priority to it and give it to the
      network)

   AILANBIOS RUN_BURST=10
---------------------------------------------------------------------

These are just recommended places to start.  There are MANY things that
effect network performance and no 2 systems are the same.

Good Luck

     >>--->  Phil  <---<<




Date: 01-31-91 (10:23) (Pvt)    Number: 3286
  To: MAX BERNARD               Refer#: 3284
From: PHIL GORDEMER               Read: YES
Subj: LANTASTIC THROUGHPUT        Conf: LANtasti

-> I saw that message and tried to follow your guidelines.  For one
-> thing, if I set my Network Buffers to anything larger than 8K, I get
-> an error message, when the LAN is initialized, complaining about
-> insufficient memory.  The same thing happens if I increase the number
-> or the size of

You must be trying to load Server High, I do NOT recommend it.  A
general statement can be made that any TSR that has a lot of hooks into
DOS and is I/O bound will suffer when loaded hi.

-> Do you know how to interpret and what to do about the various errors
-> reported by the LANCHECK program (i.e. Alignment, Collisions,
-> etc...)?

Don't get yourself hung up on Alignment and Collision errors, THEY ARE
NORMAL.  IN all networks there will be collision errors.  You only need
to be concerened when the error % count is unusually high.  Run Lancheck
on both and what you are looking for is the error indexs to be both
about the same.  The only one to keep a real eye on is CRC errors,
(again some are normal), but in almost all cases are caused by bad cable
ot bad terminators.

Thanks Phil.  Could you please explain how you arrived at those
numbers and why those particular numbers seem to work best with
PCBoard systems? I'm trying to understand how the different pieces
and options of this LAN fit together, in order to tune it for best
performance, not just with PCBoard but any other applications.  The
LANtastic manual does a pretty

Without a long disertation on LAN fundamentals, the buffers in the AE2
command line (which you should not change from default) effect the size
of the packet that gets moved across the cable.  The buffers in REDIR
are how much data is put into those packets and in Net_MGR Server
buffers is how much data to process in a single task before going to the
next task.

A simple analogy is this, you have a train that can go around the track
at a hundred miles an hour, and you have a train that goes around the
track at 50 miles an hour, More people can jump on the train that moves
slower, so more data s moved but at a slower speed.  At the 100 Miles
per hour, less people can jump on but people get their faster.

The buffers sizes come from siginificant long term testing on a number
of hardware/software platforms and my knowledge of the transfer sizes of
the information and packeting.  Remember that these numbers are not
absolute, just good starting points.  IE on my Compu-Data (22 Nodes)
system, the load is heavy so I have small buffers and lots more of them
then I usually recommend, BUT for this installation it is better.  So
how you do figure it, experimentation.  Some people keep playing with
the nubmers till they get a program like NOrton 5 Lan benchmark high,
but that does not help since those type of programs measure how well
large amounts of data gets moved.  PCB on the other hand makes very
little data in small packets.  YOu wouldn't want to allocate a buffer
that is 10 times the size of the actual data being moved, since the
system will have to send the ENTIRE buffers, not just the data..

     >>--->  Phil  <---<<
