stress and channel statistics

New Message Reply Date view Thread view Subject view Author view

Lance R. Marrou (marrou++at++vsl.ist.ucf.edu)
Tue, 20 Dec 1994 13:02:29 -0500 (EST)


Hi,

  First, I am using Performer 1.2 on an Indigo2 Extreme with IRIX 4.0.5H. I
have been working on dynamic load management and analyzing the channel
statistics. I have turned on the fast clock and fast itimers (with a call
to "ftimer -f on -i on").

My first question: is the pfClearChan supposed to take a long time (with
respect to the hardware I am using). It says in the Performer manual that
"under normal circumstances" the first segment in the statistics graph
shouldn't be visible at all. Does channel pass through data also heavily
affect this?

My second question: With a flag, I can call a GL object to throw stars (points)
up on the screen before anything else, i.e. pfDraw, so that they are in the
background. While true, the first segment is large and the time for the pfDraw
is not so big. While false, the first segment is small, but then the pfDraw
time gets much bigger (twice the usual). What could be causing this? I know
those stars should be in some performer structure so that they can be
rendered in the pfDraw function (and probably be more efficient), but I
haven't had the time yet. For now, I am happy with turning them off or on.

My third question: Load is never, I repeat never, below the low boundary of
the hysteresis band. Is this a result of the limited capability of the
Extreme? I know Performer is primarily geared for the Reality Engine, but it
seems low-end capability would get the most use out of the dynamic load
management. Also, the load fluctuates quite a lot. I assume this is due to
external processes (like NFS, et al), causing me to swap. If I lock the
code in memory, do you think this fluctuation should disappear? (just so you
know, it doesn't :) The result of the stress never dropping below the low
boundary is that the stress eventually reaches the maximum (the speed of this
depending upon the scale, of course).

Thanks in advance and may pfSanta be generous to you all. :)

_______________________________________________________
                         E-mail: marrou++at++vsl.ist.ucf.edu
IST __ WWW: http://www.vsl.ist.ucf.edu/~marrou
Visual / / ______ /\____ ______ ______
Systems / / / _ / / __ // ____// ____/
Lab / /__ / /_/ / / / / // /___ / __/_ R. Marrou
________/____//____/\\/_/ /_//_____//_____/____________


New Message Reply Date view Thread view Subject view Author view

This archive was generated by hypermail 2.0b2 on Mon Aug 10 1998 - 17:50:45 PDT

This message has been cleansed for anti-spam protection. Replace '++at++' in any mail addresses with the '@' symbol.