Google Summer of Code 2010/VNC
QEMU's built-in VNC server only supports the Hextile encoding. Hextile only uses tile-based compression. Other encodings are more popular such as Tight and ZRLE that use zlib-based compression. These encodings significantly reduce the bandwidth required by VNC.
Another side of the project is to get a threaded vnc server in order to avoid blocking guest while using heavy encodings.
- Name: Corentin Chary
- Email: email@example.com
QEMU's built-in VNC server only supports the Hextile encoding. Hextile only uses tile-based compression. Other encodings are more popular such as Tight and ZRLE that use zlib-based compression. These encodings significantly reduce the bandwidth required by VNC. This project would start by implementing support for Tight or ZRLE. Both encodings support a combination of tile-based compression and zlib-compression. Initially, one would implement zlib-compression only and compare the results to our Hextile implementation. Depending on progress, tile-based compression mechanisms (such as Tight's gradient algorithm) could be explored. At each stage, we expect that performance comparisons are done to determine the effectiveness of the additional algorithm. Additional compression mechanisms like Tight's JPEG encoding could also be evaluated.
RRE stands for rise-and-run-length encoding and as its name implies, it is essentially a two-dimensional analogue of run-length encoding. RRE-encoded rectangles arrive at the client in a form which can be rendered immediately and efﬁciently by the simplest of graphics engines. RRE is not appropriate for complex desktops, but can be useful in some situations.
Note: the CoRRE encoding is almost never useful - Hextile is a better encoding using the same ideas. CoRRE (Compact RRE) is a variant of RRE, where we guarantee that the largest rect- angle sent is no more than 255x255 pixels. A server which wants to send a rectangle larger than this simply splits it up and sends several smaller RFB rectangles.
ZRLE stands for Zlib1 Run-Length Encoding, and combines zlib compression, tiling, palettisation and run-length encoding. On the wire, the rectangle begins with a 4-byte length ﬁeld, and is followed by that many bytes of zlib-compressed data. A single zlib “stream” object is used for a given RFB protocol connection, so that ZRLE rectangles must be encoded and decoded strictly in order.
We'll need something to compare different encodings and how much bandwidth/cpu they use. For that we can probably use http://www.lesswatts.org/projects/bltk/ with Idle, Office and Dvd Player workloads.
Support for tight is complete. JPEG and gradient should be merged soon.
This is a experiment to speed-up noVNC (http://github.com/kanaka/noVNC/ ). This adds a new encoding similar to tight (but without basic/zlib compression): Tight PNG encoding.
Threaded Encoding Queue
This project aims to create a queue for encoding computations.
If you build qemu with --enable-vnc-thread, the queue will be threaded and (hopefully) send_framebuffer_update() won't block the guest execution anymore. You will still be able to use a synchroneous queue (current behavior) with --disable-vnc-thread.
Video Detection and specific encoding
Adaptive Tight encoding
ZRLE // ZYWRLE