Duane Percox ([log in to unmask]) wrote:
 
: [log in to unmask] (Jeff Kell ) wrote:
: [ major snipping...]
 
: [Climb on SoapBox]
 
  I know you all will see me very biased (since I'm the NS/VT lab engineer
  at HP CSY lab since May'94)...
 
  However, I'd like to bring it to your attention that given my background
  of supporting NS products (including VT) in the field and in various HP's
  international support centers and response centers since 1986 (yes! scary
  nine(9) years!!) - I have seen many systems with high or very high VT usage.
 
  The biggest ones at customer sites I've seen to date have been 992/400
  CPU's with 1.5..2 GB of main memory running with daytime 'normal' load
  of ~900 sessions of which about 700 VT and another 100 other NS servers.
  I was amazed by the response time on that system - felt like I was the
  sole user on it!!
 
  In benchmarking situations here at HP I was equally amazed when I saw
  a 4-way KittyHawk system running some oracle(?) database benchmark with
  over 1000 VT sessions driving it... it seemed to be very lightly loaded
  what I could tell from the response time.  Test drivers were about half-
  dozen (or more, don't know the details) 867 class CPU's driving scripts
  through vt3k sessions.
 
  Also, I frequently see systems with 300..500 VT sessions and it does not
  seem to be any problem for 99x class CPU's providing they have adequate
  memory supply.  As such, I would expect high end 9x7 CPU's to be quite
  happy with 200..300 VT sessions.
 
  It has been correctly stated that if switching from 200..300 DTC sessions
  to same number of VT sessions, your systems' resource utilization will go
  up with the same number of additional VTSERVER processes.  Assuming 200
  processes for the system + 300 users with some process handling and HPDESK
  type jobs with multiple children - averaging about 5 processes per job/sess
  you'd be talking about 200+5*300 = 1700 processes currently running on your
  system at peak times.  Adding to that one VTSERVER process per session
  would bring you up to 2000 processes - 18% increase in number of processes,
  and of course increase in transient space and memory utilization.
 
  What comes to the additional terminal i/o overhead one should keep in mind
  that no application is 100% terminal I/O.  If your applications CPU usage
  currently amounts to 70% of the total CPU available - and the CPU consumed
  in applications' terminal I/O is a total of 10% of the applications' CPU
  (rest goes to database access, file system etc...) - this means 10% of the
  70% - i.e. 7% total system available CPU used for terminal I/O.
 
  I often use a figure (although have no exact number to give) of 2..3 times
  more overhead for VT than DTC terminal i/o... using that in above example
  would mean that the additional CPU needed for VT would be in the range of
  7..15% more CPU than you're currently using.  Again, this is mainly
  guesswork as I don't know your applications and how big part of your
  applications mix is CPU needed for plain terminal i/o...
 
  Hope this helps in your planning,
  Best regards  :-),
  Eero Laurila - HP CSY Networking lab, NS services, VT.