Re: very slow drawing

From: Thomas Walter (twalter@physik.unizh.ch)
Date: Sat Feb 06 1999 - 09:15:19 MET


Rene,

> > > > my root-macro executes very fast until it reaches the part,
> > > > where the postscript file is generated. The cpu-usage drops
> > > > from about 99% to less then 3%.
> > > >
> > >
> > > Thomas,
> > > In this part of the macro, you draw 400 histograms. This may take a long
> > > time
> > > depending how many channels you have per histogram and your options.
> > 
> > 50*( 3 * 128 channels + 2 * 100 channels + 10 ch + 25 ch + 50 ch)
> > 
> > > Is your Postcript file generated on an AFS disk?
> > 
> > No it is generated on the local scratch disk.
> > 
> > > What is the size of the generated ps file?
> > 
> > 2973 KB
> > 
> > But all this doesn't explain the very low cpu-usage. One further thing we
> > have to consider the display. I'm computing in Zuerich and displaying
> > in Hamburg over a ssh-connection.
> > 
> > Is it possible, that root is waiting for some X11-messages? If this is the
> > case, is it possible to generate the ps-file without display the
> > histograms on X11?
> > 
> 
>  Thomas,
> It is clear that you have a problem with the line. The time is not taken
> by Postscript but by X.
> You can generate the Postscript file by running Root with the batch
> option
>   root -b
>   Root > .x macro.C
> 
> Let me know the result.
> 
Here they are:

Part1 of my macro fills the histograms, Part2 does some fits and in Part3
the histograms are displayed and the ps-file is generated.

These are the results for interactiv root. See the big differencies
between the RealTime and the CPUTime in Part 3:

Part 1  RealTime: 82.29  CPUTime: 79.39
Part 2  RealTime: 1.97  CPUTime: 1.93
Part 3  RealTime: 536.55  CPUTime: 36.53


And here the results gained with root -b:

Part 1  RealTime: 81.3  CPUTime: 79.8
Part 2  RealTime: 1.99  CPUTime: 1.95
Part 3  RealTime: 5.5  CPUTime: 4.6

As you can see, the second version is a factor 10 faster than the first
one.

If I remember me correctly, this discrepancy already existed in Zuerich.
I displayed the histograms over the ethernet-lan from one computer to
another. So I guess, the problem isn't a slow link between Hamburg and
Zuerich.

Thomas Walter



This archive was generated by hypermail 2b29 : Tue Jan 04 2000 - 00:43:29 MET