Fons Rademakers writes:
>
> - The thing that slows down ROOT graphics is that we (still) use (too)
> many round-trip X commands (querying pixmap attributes, cursor shape
> changes when moving the mouse, etc.). It is not so much the amount of
> data that is transfered as well a latency problem (in spite of the
> X command buffering).
>
Hi Fons,
thanks for the detailed explanation. It seems that I've been
misinterpreting the actual reasons of the slow graphics - sorry for that.
However it is in our common interests to figure what the actual reasons are and
to fix them. THinking this way I've done a small test which agrees with
what you're saying. I started ROOT and opened on the screen 2 canvases -
one 500x500 and another one 700x700 pixels.
Here is the timing (time necessary for PAWX11 to open a graphics window and to
get to the command prompt is given for the comparison) :
canvas on the screen drawing complete (all the menu items poped up)
500x500 10 sec 30 sec
700x700 10 sec 30 sec
PAWX11 5 sec
So you see that
a) time doesn't depend on the window size - once again my apologies for
the confusion
b) I have *really* sensitive tools to study graphics performance in real time
(a remote X-terminal with 28K modem connection).
c) it takes 20 seconds to display a menu bar ("File" , "Edit" ... "Help")
Accuracy of the measurements is about +/- 1 second. Then it takes about 5
seconds to draw an empty histogram:
TH1F* hist = new TH1F("a","a",100,0,1)
hist->Draw()
and about the same time (5-6 sec) to plot non-empty histogram:
for (int i=0;i<100;i++) hist->SetBinContent(i,i)
hist->Draw()
so everything looks like you're right in the understanding of where the
time goes. I'm wondering if anybody from people working on event displays
would be interested to see where it is possible to minimize the number of
round trip X-commands and thus improve the over-network performance.
-pasha
This archive was generated by hypermail 2b29 : Tue Jan 04 2000 - 00:43:29 MET