Re: [dvd::rip] Cluster Mode System Requirements
|
Subject: |
Re: [dvd::rip] Cluster Mode System Requirements |
|
From: |
Jörn Reder <joern@xxxxxx> |
|
Date: |
Thu, 4 Aug 2005 09:30:10 +0200 |
Phillip Geiger wrote:
> Am I wasting my time with a bunch of 333 MHz dinosaurs?
Difficult to answer this question... I'd say you could give it a try,
because of the high number of "333 MHz dinosaurs" ;) If you decrease the
chunk size to get at least 10 video chunks they all should be busy with
"their" chunk. Your fast machines should do the audio / merging jobs
(which take a bit longer, because they're not chunked like the video).
You achieve this by telling dvd::rip that the dvd::rip data harddrive is
locally connected to your fast nodes (even if that's not true; it's just
a hint for dvd::rip to put these I/O relevant jobs on such a node).
The problem is: dvd::rip's cluster scheduler currently isn't so smart to
not giving a job to a slow machine (it even don't know which machines
are slow), if a fast machine would be better for this job, but the fast
machine is currently busy with another job, which may be finished in a
few seconds. So the slow machine starts with the job, while the fast
machine is idle after finishing its current job (was that
understandable? difficult to explain such things as a non-native english
speaker... ;)
But I'm planning to extent the cluster node configuration that way.
Probably an index value telling dvd::rip how fast this machine is, so
dvd::rip could try to prevent such situations.
Regards,
Joern
--
Joern Reder
Development Head ZYN! Coding Division - http://www.zyn.de/
pgp00008.pgp
Description: PGP signature
|
|