[vlc] Re: variable CPU usage

[Michael S. Zick]

On Sun June 6 2004 21:32, Abhinav jalan wrote:
> Hi,
>
> Thanks for replying to my previous query.
>
> I am trying with an experimenting of transcoding a stream on the network.
> For that I need to find CPU usage of the transcoder for a particular
> configuration. When I use http as source, for each experiment, CPU usage
> comes out to be same. But when I use a streaming source (udp), I do not get
> consistent reading for CPU usage. And sometimes the variation is of the
> order of magnitude. However, within a single experiment, variation within 
> the CPU usage does not vary much as the transcoder is transcoding a stream.
>
> In an another experiment, I made the streaming server to multicast a
> stream. And made 8 nodes to transcode the stream with same configuration.
> All the 8 workstation had same hardware configuration. But I got different
> CPU usage on each node.
You have chosen a very difficult experiment to setup such that repeatable
measurements can be expected.
Variables include many operational details beyond 'same hardware
configuration'.
The operating systems on which VLC runs are 'multi-process' - the hardware
is time-shared between the multiple threads of VLC and everything else that
is executable at any one moment.

>
> Do you know the reason why I do not get consistent readings. Is there a way
> to stabilize the CPU usage of the trancoder when the source is a streaming
> server.
I would suggest that you first examine your experimental setup - make it as
'clean' as possible and identical among all machines.
As an example (Linux):
1) boot each machine with the kernel command line including: init=/bin/bash
2) start only services required to access the network connection on your setup
3) run VLC at a real time priority rather than the default, demand adaptive,
scheduling.
4) examine carefully how you are measuring CPU usage.
Questions:
A) Are the CPU usages relatively similar now between nodes?
B) Is the difference between HTTP and UDP now consistently repeatable?

I hope these hints will help you with your experiment.
Mike
>
> Thanks,
> Abhinav

-- 
This is the vlc mailing-list, see http://www.videolan.org/vlc/
To unsubscribe, please read http://www.videolan.org/support/lists.html
If you are in trouble, please contact <postmaster at videolan.org>