Difference between revisions of "VLC Dev Projects"
| Line 33: | Line 33: | ||
There are many bottlenecks currently limiting the performance and scalability of VLC as a streaming server: | There are many bottlenecks currently limiting the performance and scalability of VLC as a streaming server: | ||
| − | + | * The I/O blocks are linear, which implies lots of avoidable memory copying and dynamic memory allocations (particularly in the streaming output). | |
| − | + | * mdate() (the timestamping function) is invoked in too many places to count them; but it is not a cheap operation on modern systems and should be avoided on the fast path. Ideally, a single thread would only need to call mdate() at most once per wake-up. In some case, mdate() is only used for debugging purpose which is really inefficient. | |
| − | + | * The HTTP/RTSP core is single-threaded. This prevents scaling to SMP systems (which are the norm on server-side nowadays) properly. This also implies not very scalable I/O event polling. Finally, this gets problematic if one HTTP/RTSP client triggers a computationally intensive operation (such as establishing a TLS context if HTTP/SSL is used). | |
| − | + | * There are many dynamic memory allocations that could be avoided. Video filters and codecs are very careful about this, but the stream output plugins are not. In many case, a stack-based buffer can be used (stack buffer, has constant time fast lock-less allocation, and needs not be freed explicitly) instead. | |
Other improvements that would be particularly benefitial on the server side: | Other improvements that would be particularly benefitial on the server side: | ||
| − | + | * Memory allocation failure are not always duly handled. In some cases, that could potentially be leveraged by a client to crash a VLC server by causing a very large allocation attempt. | |
| − | + | * Rewrite and revive the root wrapper to allow VLC to bind to the RTSP (554) port while not retaining root privileges. | |
Contact: [[User:Courmisch|Courmisch]] | Contact: [[User:Courmisch|Courmisch]] | ||
Revision as of 18:38, 6 March 2007
This is a list of small independent improvements to VLC for which we are looking for interested developers, because we simply don't have enough time.
These do not require advanced knowledge of VLC internals but they can still vastly improve many parts of VLC
For each project, we list a contact person, who knows the subject best and who will be able to assist you if you are interested in working on these. We generally are very often present on irc.videolan.org, chan #videolan, so please don't hesitate to drop by or to send a mail if you are interested in a given problem (our mail adresses are generally <nick>@videolan.org).
We also have a number of "easy" bugs. You can check them out here
Easy projects
Video Filters
You can do video filters for VLC media player.
You can inspire yourself on this list of filters from a commercial product.
You can start the code here.
Subtitles improvements
VLC media player does not support all the styles for a lot of subtitles type. Take a look at here to start coding.
Network
RTCP support
Our RTSP implementation (for broadcasting and video on demand) does not support the RTCP protocol. This is a must to have a real support for RTSP. Some work has already been started, but there is still much to do.
Contact: JPSaman
Performant VLC streaming server
There are many bottlenecks currently limiting the performance and scalability of VLC as a streaming server:
- The I/O blocks are linear, which implies lots of avoidable memory copying and dynamic memory allocations (particularly in the streaming output).
- mdate() (the timestamping function) is invoked in too many places to count them; but it is not a cheap operation on modern systems and should be avoided on the fast path. Ideally, a single thread would only need to call mdate() at most once per wake-up. In some case, mdate() is only used for debugging purpose which is really inefficient.
- The HTTP/RTSP core is single-threaded. This prevents scaling to SMP systems (which are the norm on server-side nowadays) properly. This also implies not very scalable I/O event polling. Finally, this gets problematic if one HTTP/RTSP client triggers a computationally intensive operation (such as establishing a TLS context if HTTP/SSL is used).
- There are many dynamic memory allocations that could be avoided. Video filters and codecs are very careful about this, but the stream output plugins are not. In many case, a stack-based buffer can be used (stack buffer, has constant time fast lock-less allocation, and needs not be freed explicitly) instead.
Other improvements that would be particularly benefitial on the server side:
- Memory allocation failure are not always duly handled. In some cases, that could potentially be leveraged by a client to crash a VLC server by causing a very large allocation attempt.
- Rewrite and revive the root wrapper to allow VLC to bind to the RTSP (554) port while not retaining root privileges.
Contact: Courmisch
Inputs
DVB support for Windows
DVB is the norm for :
- Digital Terrestrial TV (aka DVB-T)
- Digital Cable TV (aka DVB-C)
- Digital Satellite TV (aka DVB-S)
Several DVB input cards exist. We support these on Linux, but not on Windows. For Windows, we need to support so-called BDA drivers.
The problem is that we don't have both the time and the required hardware on Windows comps.
This project is probably not terribly complex, but you need the hardware. Someone has already began it on the forum, so start from there.
Contacts:
V4L2 support
VLC currently has a pvr module and a v4l (video for linux) module. However
- some cards are only supported in the new v4l2 API
- The linux kernel will soon drop v4l support
So if someone could update VLC's current modules to work with V4L2 and the new ivtv api, this would be a welcome addition. This is not a very difficult project, but you will need to have the hardware to test against. The code has been began, so it should not be too difficult.
Video
DirectX 9 video output
DirectX 9 features some improvements that we could use in our video output
Contact: Quovodis
Improvements to the X11 video output
Fullscreen support to improve .
Interfaces
Streaming profiles
This is a big subject, that has barely started. Interested persons will need to work closely with existing developers to implement this. Not much information is available yet. The idea is to make streaming very easy, and yet powerful by using parametrable profiles. These could be used either by the graphical interfaces or through HTTP/command line interfaces
Contact: Zorglub
Playlist
Improved Podcast support
Our current podcast support is very weak. We can read them, but that's all. We need a real infrastructure to handle them correctly. This will require some changes to the playlist code, so you will need to work closely with existing developers.
Improved ASX support
ASX is a playlist format used by Microsoft stuff. We can read it but we could do much better. More information is available at https://trac.videolan.org/vlc/ticket/20
Contact: Dnumgis
Disc probing
We currently have some miscellaneous code to detect discs (DVDs, Audio CDs, ...) but it's a bit scattered and it could be used much better in the playlist.
Contact: Zorglub
Live folders
Some programs support so-called live folders. You add a folder to the playlist, and if the files in the folder are changed, the changes are copied to the playlist.
Having this would of course be great :)
It might require some work to be able to do this in Linux, Windows and OS X, but the current developers will of course be able to help on these.
Misc
DBus Integration
DBus is a communication mechanism for Linux (mainly).
We would like VLC to be able to use it so that other programs can interact with vlc (start/stop, know what is being played, ...)
Work is well advanced, see DBus
