Re: [OpenSTA-users] Severe load ramp-up limitation encountered

[Daniel S. <da...@Op...>]

Bernie Velivis wrote:
> Aside from the synchronize requests appearing after the end timer
> for SP01_load_page, there is nothing in the script that looks
> suspicous to me.

I'm never sure about the SYNCHRONIZE at the end of a script - I
wonder if SCL flow just ends without any SYNCHRONIZE whether the 
connections just get dropped even if they haven't completed or
whether, in fact, there is kind of automatic SYNCHRONIZE when a
script comes to an end - ie. the script doesn't actually end until
all of its active connections have finished doing there stuff ...
and what does this mean if you are using HTTP keep-alive?  Will it
wait until connection timeout? or stop when active HTTP requests
are all done?

> Perhaps you are assuming the disconnect 1 command is a synchronize
> command?  I dont think it is.

According to the documentation a DISCONNECT <conid> will actually
wait until the HTTP requests on the connection have finished before
closing the TCP connection - so in one sense the DISCONNECT does
act like a SYNCHRONIZE for a single connection.  If you do a
DISCONNECT ALL then that would be like a SYNCHRONIZE followed by
clowing all the TCP connections.

Don Downing wrote:
> > Now, as to my installation issue on my XP Pro/SP2 laptop:
> > Modeler and name server work fine, but I cannot open a Test,
> > and the icon for the Test in the Commander Explorer appears as a
> > "Windows Generic" icon rather than the red-blue-green clover
> > icon.  I just deinstalled/reinstalled.  Figure I must have a
> > conflict with some other app/dll...Ever encounter this?

I think this is just a Windows-ism, something which actually seemed
to get better in XP - who knows what causes it but it is a sign of
unpleasant things happening in your registry ...

> Also, I tried your script verbatim, and saw longer response times
> (makes sense, my original had only 1 get, again more or less linear
> to 500 users but sp01_load_page timer ranged from .96 to 40 seconds.

So this might be a Windows 2000 Server thing - although I don't
remember anyone reporting potential issues with this OS before,
I remember 2003 Server gripes that we never got to the bottom of
but not W2K Server to my recollection.

> Minus the network differences, this seems to indicate a problem
> unique to your load server.

Or something that is being masked by your more limited intermediate
network - we really need some tests in a fast switched LAN
environment to rule out this potential explanation.  It does however
seem much more likely that this is an OS version related thing.

Cheers
/dan
-- 
Daniel Sutcliffe <Da...@Op...>
  OpenSTA part-time caretaker - http://OpenSTA.org/