On Thu, 2009-07-30 at 10:53 +0200, Jörn Reder wrote: Hi,
> > I'd like to remark that > > 1) PVM/clustering support will RE-ENTER someday in the future > > 2) I'm more than happy to support in any way I can anyone want to > > develop, introduce or mantain that code. > I never used the PVM stuff and don't know how it works exactly. But I > think the way one can do cluster stuff today is sufficient: splitting > the stream (logically) into independent chunks and process chunks on > different nodes sharing the storage via NFS. Later merge the transcoded > chunks back into one target file. I do agree. IMO, this basic design works well enough, and it is also simple and relatively easy to implement, and that's a good thing. My pet peeve is to use something different from NFS (maybe a transcode frame server or something like that), but that's a very minor aspect. With the PVM3 code, we had two problems: 1) no active developer used regularly that code; it was effectively unmantained and, very likely, just broken. 2) given the recent developments into PC world, IMO single, multi-core boxes are cheaper and more cost-effective than clusters. AKA: clusters (for the average Joe User) are becoming even less common than before. [...] > So I think improving transcodes navigation/splitting and A/V sync > capabilities would be more worth than adding PVM stuff again, since it > would be useful in single and cluster mode. I wholeheartly agree. The next development cycle (1.3.0) will start to fight this problem by redesigning the import layer like we did for the export layer during the past two (1.1.0 and upcoming 1.2.0) cycles. This imply serious structural changes like (in no particular order) 1. introduction of real timestamps and time accounting into transcode 2. more powerful synchronization engine 3. introduction of a demuxer thread (open the input just once, not twice - welcome real DVB support!) 4. move the filters into decoder thread from the encoder one 5. deep review of how probing works 6. (maybe) review of the overall navigation architecture We'll discuss all of them, and even more, into the (hopefully) near future, after 1.2.0 released :) Bests, -- Francesco Romani // Ikitt http://fromani.exit1.org ::: transcode homepage http://tcforge.berlios.de ::: transcode experimental forge
