-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 leandro ribeiro schrieb: > So my questions are: ... What on earth is a Low Latency or Realtime > kernel and why should I cross the street to say hello to it?
Hi Leandro, as the previous posters already pointed out: you don't need a RT kernel for video work and esp. for Cinelerra. To fill in some informations: By "latency" we denote the time the system needs to react. For example, when an event (like a keypress) or an input signal has to trigger some reaction. Contrary to analogue systems, where latency is mostly limited by the bounds of physics, digital systems do programmatic processing of signals and events. Thus we need to store the data into a buffer, process it and then send it out again. Here, the size of this buffer is the factor determining the latency of the system. So, ideally you want to make the buffers as small as possible, but your system needs to be fast enough to cope with such small buffers. While basically most modern systems are fast enough, the problem is rather that the typical operating system kernel was not designed for such operation; just as the system events come in, the kernel may spent some irregular amount of time to handle its internal bookkeeping duties, e.g. organizing the files on disk. Because, for normal desktop use, it doesn't matter if there is a delay of 50ms. Sadly enough, for audio work, when there is possibly sometimes a pause of 50ms caused by the kernel organizing internal stuff, this would force us to make the buffers large enough to cover this gap, which would mean we get a latency > 50ms. But a delay of even 20ms is a problem when playing music. Thus, Ingo Molnar and other Kernel Hackers started a series of patches reworking the kernel infrastructure such as to make most kernel operations interruptible on a small time scale. Parts of this work have been accepted into the mainstream kernel, while other parts can cause stability and/or security problems and are deemed experimental. Generally speaking, reducing the /latency/ of operations within the kernel doesn't come for free, it slightly /reduces/ the throughput, i.e. the bare speed of the system. So, generally it's a tradeoff: do you need maximum speed, or do you need the system to reliably react within a certain timespan? Because all of this is complicated stuff and not every realtime patched kernel version is equaly usable in practice, you should always rely on the pre-built RT-kernel images which are provided by specialized distros like "Ubuntu studio" or "64 Studio" Cheers Hermann V. -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFIvzb2ZbZrB6HelLIRAhLPAKDfTgSyuzdl1C8h4zUWX/vJH2C9hwCdF8o7 IAsO0+MndU2Zzw0fJy5Zmik= =XisS -----END PGP SIGNATURE----- _______________________________________________ Cinelerra mailing list Cinelerra@skolelinux.no https://init.linpro.no/mailman/skolelinux.no/listinfo/cinelerra