Excerpted from a message dated 04-16-98, Peter French to Murray Lesser:
PF>  In this case, I can't see how it is possible to get at the
  >benchmark applet without being "connected".  If I could download the
  >sucker, I'd try the JVM "native".

ML>     If you mean what I think you mean when you said that you ran the
ML> tests "while connected" you had the internet transmission delays
ML> involved, as well as the browser overhead.

JF>  Thought about that, and tried it in the early AM when the US is
  >asleep - and got very similar results.  I assume that the applet is
  >downloaded - and then run in my system - where the measurement is
  >taken, so network did not make much difference.
Hi Peter--
    As I warned in my earlier message, I don't know much about Java
(although I have played a little with NetRexx).  But it is my
understanding that if you are running an Applet from the Web, your
system keeps getting the next method from the source, which means there
has to be a turnaround to ask for it, and then another turnaround to get
it, and then a security check to make sure it isn't dangerous, for each
method used.  (I may be wrong about this.)  It seems very strange that a
"benchmark" should be written as an Applet and not as an application.  I
would assume that JStreet Mailer is a Java application, not an Applet.
PF>  My system still does not measure up to some of the numbers I've
  >seen, but that may be the hardware design...
  >                    ... My problem was, that I had followed some
  >advice blindly, and disabled the JIT compiler in my CONFIG.SYS by the
  >command SET JAVA_COMPILER=XXX
    You can "reinstall" the JIT by removing that line in CONFIG.SYS.  It
isn't disabled; you have just told your system the wrong place to look
for it.
PF>  Of course, why I was looking into all of this was to try and speed
  >up the Java performance of applications such as JStreet Mailer. 
  >Turns out, that the JIT compiler was being turned on by the setup I
  >had for JStreet anyway, so I've seen no improvements there.
    I don't understand how the JStreet Mailer can override CONFIG.SYS,
but I'll take your word for it.  I hope you are not running the Mailer
under Netscape!
    "Initializing" a Java application is very slow, perhaps because it
is essentially a "load and compile" operation.  Once loaded and
compiled, running under the JIT is supposed to be fairly fast, but not
as fast as an equivalent application compiled from an "early binding"
language.  I noticed in my playing around with Applets written in
NetRexx (running them under APPLET.EXE, not under Netscape) that the
initialization was slow, but subsequent runs of that same small Applet
(without closing it in between) were very rapid.  Did you try
"reloading" your benchmark Applets without closing them in between?
    I have now run out of suggestions other than to forget benchmarks.
    Regards,
        --Murray

___
 * MR/2 2.25 #120 * If you are not confused, you don't understand the 
situation
--- Maximus/2 2.02
---------------