[clug] remote X performannce
hugh.fisher at anu.edu.au
Mon Aug 3 18:29:56 MDT 2009
Michael James wrote:
> Running from a machine on the same local switch
> we get visible re-draw delays but usable performance.
> Running from Brisbane back to Canberra we get ...
> "You have re-sized your screen, go have lunch
> while you wait for the changes to take effect".
> The network looks good, a long ping gives
> no packet loss and a rock steady 16.6 mSecond delay.
> Why is the network having such a bad effect?
The application coders might not have considered or tested
remote display over a network. A number of X API calls
are round-trip requests which will introduce a delay equal
to the network latency. For instance, if the GUI kit has
toolbars with tooltips and uses XQueryPointer to test if
it should be drawing a tip over a button, that's a round
trip delay in the rendering. Another bit of code queries
which window has the focus for keyboard events so it can
draw a special border around it ... it won't take many
of these to generate the behaviour you're seeing.
(Yes, the code shouldn't have been written that way, but
most people, myself included, don't regularly test their
programs over a WLAN.)
I'd suggest trying NX, or if that doesn't work, consider
VNC. No, really. VNC takes the brute-force approach of
just shuttling blocks of pixels around, so won't be
affected by any strange interactions between the app
code and X.
More information about the linux