Min Sys Req's and CPU

[Gustin Johnson]

Kiernan Holland wrote:
> I second this.. BTW, is hyper-threading being on a bad idea?

Depends on the kernel version and the hardware.  In my experience recent
kernels (2.6.28+) seem to handle HT better.
> 
> I'm visiting my brother who works for Hughes Research lab, and does work
> with linux and FreeBSD all the time. He says that one of the problems
> with realtime kernels is the potential to have cores competing for
> resources, resulting in dead lock. I read that this happens in
> processors such as the Pentium 4 with hyper threading, as it tends to

The P4 was terrible all around.  The whole point of the netburst
architecture was to beat their rivals with Mhz instead of building
genuinely superior hardware.  It seems Intel is back on track again.

> starve one thread of resources (in the interest of making maximum use of
> both processors, also termed as cache thrashing. Also there is some

Most of the default kernels do not support hyper threading.  This is one
of the reasons I build my own kernels, there is some significant
performance gains when I enable hyper threading on my new Core2 Duo.
The more recent kernels seem to actually handle HT properly, not as a
discreet core which they used to (an actually caused a decrease in
performance in some situations.

<snip>

> I recall in my CS classes that if you force processes to allocate
> resources by the same method that is used in post offices to order
> people into lines without literally forming a line (by taking a number),
> you can avoid greedy resource allocation that leads to deadlock by
> resource holding. I don't know what this method is, but it seemed
> convincing to me as a way of avoiding these lock-ups at the time.

If you really care about the scehduling, there is a lot of info in the
Linux Kernel Mailing List, there has been a lot of work on the
schedulers over the years.  I suspect that the answers to a number of
your questions can be found in the source code itself.
> 
> I know this sounds like I'm searching for a answer to this Xrun problem,
> but the solution may be better to suggest a way of working when making
> compositions with Jack and the other audio applications, like say being
> able to layer (or automate the layering) of tracks using rosegarden and
> ardour. I know some people care more about performance with Jack, but
> with the use of ardour, it would be possible to complete work with
> poorly performing audio applications, like ZynAddSubFx which can bring a
> CPU to it's knees quite easily. I'm personally stuck on trying to get
> rosegarden to play and ardour to record in synchronization. I've tried

This works fine for me.  I have midi in Rosegarden that gets passed to
either zynaddsubfx, ams, linux sampler, or an outboard synth.  The audio
is then passed into a track in ardour.

> lots of things with no success.. If I find the solution, I'll post it on
> youtube where others can get the answer without having to come here and
> plug the listserv for information. One way to reduce questions is to
> make information resources more accessible. A listserv is nice in one
> way of providing help but it isn't very effective at reducing the
> frequency of common questions, a FAQ isn't very helpful, how about a
> video FAQ?
> 
FAQs are only not useful when people do not read them.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 260 bytes
Desc: OpenPGP digital signature
Url : https://lists.ubuntu.com/archives/ubuntu-studio-users/attachments/20090907/d59ae586/attachment.pgp