Understanding noVNC performance.

1,274 views
Skip to first unread message

Miguel Moll

unread,
May 29, 2013, 3:21:16 PM5/29/13
to no...@googlegroups.com
I've been testing noVNC and messing around with it. Very impressed with the tech involved. After doing some reading, especially https://github.com/kanaka/noVNC/wiki/Performance-notes even though it might be outdated, I was wondering if the conclusion I came to is correct:

Unless I use a VNC server that supports tightPNG, noVNC will be slow (because of zlib in JS == bad)?

I have been running noVNC on Ubuntu 12.04 connecting with Chrome (latest) to the vnc server that comes by default with OSX 10.7. When I connect with a "regular" vnc client, interaction with the GUI is spot on and usable.

Maybe someone can correct me? Thank you!

- Miguel

Joel Martin

unread,
Jun 6, 2013, 11:14:25 AM6/6/13
to no...@googlegroups.com
Miguel,

It turns out that in most cases the performance of the standard tight encoding is actually quite good. The advantage of tightPNG is that it requires less CPU for the client (because it can mostly just blit the data to the canvas without much decoding). You won't get native performance with noVNC (at least not until a full conversion to binary data + asm.js code happens). There is a patch mentioned in https://github.com/kanaka/noVNC/issues/221 that you might want to try. The caveat is that it may cause slower clients to become overwhelmed.

And yes, I do need to update the performance page as it is significanly out of date.

Joel Martin (kanaka)
Reply all
Reply to author
Forward
0 new messages