David Randall ([log in to unmask]) wrote:
: Eero, ref your .02,.....
 
: >>  1.  When running a terminal connection using serial line and a DTC vs.
: >>      a session through NS/VT, couple of things about the speed:
 
: >>       - True, NS/VT session consumes some more resources on the hp3k
: >>         (one more process).  Session start-up (getting the logon prompt
: >>         and session built up) is slower than thru a DTC port.
 
: ... Understood but how do we find out how much overhead? - If I move 20 users
: from DTC to network will I see a significant difference in performance on a
: fairly busy 957RX (for example) - Where in the copious documentation do I
find
: this jewel of information.
 
- Shame on me not saving the information I once typed up in this same
  newsgroup for the same very question.  The bottom line is that there are
  no general numbers to give - every system runs a different combination of
  applications doing terminal I/O differently - and it's the terminal I/O
  part of the total CPU used that really counts when comparing the two.
 
  For session start-up, you can always eliminate most of the session
  creation overhead by allocating a certain number of pre-created vtserver
  processes through :NSCONTROL SERVER=VTSERVER,min,max command.  By doing
  that the connection establishment and logging on is not much different
  from DTC attached terminals.
 
- Once the session is up and running, and the application is up and running,
  one should be able to figure out how much of the total CPU used (say, your
  CPU is 80% busy) is spent purely on the applications DTC terminal I/O.  I.e.
  exclude everything else - calculations done, data manipulation done, database
  accesses, any other filesystem access - anything the computer does but
  terminal I/O - it's only that part that gets heavier when moving from DTC's
  to VT connections.
 
  If the CPU needed for terminal I/O (out of total 80%) is very little
  - say 5% - you probably won't notice much of a difference in CPU usage
  and/or response time and overall performance.  However, if the application
  is very terminal intensive, uses lots of character mode, no or next to no
  block mode - total CPU burnt on terminal I/O is 50% out of total 80% -
  going to VT connections would probably have a pretty dramatic impact.
 
  I have no numbers to tell 100% sure, but I generally use a figure of a
  terminal I/O transaction being 2..3 timer heavier through a VT session
  as opposed to a DTC session.  In the 5% example case this would mean now
  10..15% of total CPU used in terminal I/O - total CPU usage now climbing
  from 80% to 90-95% range.  If I just had the time and permission to do
  so, I'd be interested in running some benchmarks to get some numbers
  out.  Then again, it would be user application dependent thing how much
  each different system would be impacted.  Sigh...
 
  Again, there are no rules to give that would apply to every (or not even
  most) situations - however, it helps to realize that the terminal I/O is
  the part we need to look at.  Maybe you (or someone at your site) know
  your application architecture well enough to make some guesses about it's
  terminal I/O intensivity... that would help you to make a better educated
  guess than I can give.
 
  Anyway, I guess I could add that most systems use NS/VT very extensively
  and the biggest configurations I have seen running are well over 1000
  simultaneous VT sessions - we simulated about 2000 active users on some
  kittyhawks while ago running some benchmark through VT sessions.
 
 
Comments welcome,
Cheers  :-)  Eero.  HP CSY Networking lab, NS services, VT.