Random performance problem (Linux -> W95)

[Quentin Stafford-Fraser]

Chris -

I'm surprised your timings have such wide variations.  I've certainly never
seen *anything* take 52 seconds to draw on an ethernet.  The CPU may be
lightly loaded, but hows the memory?  Are you sure that neither of them are
swapping?

There will be some differences because the VNC server sends updates when the
client has finished drawing the last set of changes.  A small change in the
timings can mean that things are batched in a completely different way.  As
the network and processor speed increases, the updates will be requested
more often, and so more data will usually be sent.  This makes direct
comparisons difficult.  Also, some applications are very inefficient about
the way they draw their displays.  Early versions of Netscape, for example,
when scrolling would redraw everything twice with each new scroll position.
If your network and client were fast, you'd get both redraws because the
update would happen frequently.  If you run the viewer with the -delay
option you can cause a black box to be drawn for the specified number of
milliseconds whenever an update happens before the update is drawn.  This is
quite fun - try vncviewer -delay 150 for a start.  (Actually on the Windows
version, it's whenever a Paint message is processed, which is almost the
same thing.)

Anyway, none of this is actually explaining your delays.  If you can test a
similar situation with a Linux viewer, I would be interested in the results.
Do you see the view being drawn very slowly, or does it appear all at once
but take a long time?

Regards,
Quentin
----
Dr Quentin Stafford-Fraser
The Olivetti & Oracle Research Lab
http://www.orl.co.uk/~qsf


-----Original Message-----
From: Chris Thoday <chris "at" thoday.demon.co.uk>
To: VNC List <vnc-list "at" orl.co.uk>
Date: 18 April 1998 13:55
Subject: Random performance problem (Linux -> W95)


>I have a CAD application running under Linux which experiences random
>delays when displayed on a W95 box. The time taken to display the same
>engineering drawing varies from 2 seconds to 52 seconds: 3, 50, 50, 4,
>50, 49, 51, 3, 3, 50, 2, 2, 5, 2, 50, 3, 2, 5, 3, 52, 50, 3, 3, 38, 2,
>2, 53, 3, 3, 3, 49.
>
>Each test was run with the same drawing difinition file and from the
>same process.
>The Linux box has a 200 MHz cpu with 32 Mb memory and the W95 box has a
>133 MHz box with the same memory. The netstat command showed no ethernet
>problems and I have experienced no delays with FTP or Samba. Monitoring
>on the Linux box appears to show only low cpu usage during the long
>delays with the interrupt 8 light flashing.
>
>Eur Ing Chris Thoday,
>Software Engineer, Rugby
>
>
>
>