Disable kscand/Normal?

Disable kscand/Normal?

[HOLTZ, CORBIN L. (JSC-ER) (LM)]

Hello,

Sorry to post to the list without being subscribed, but I've searched the
web for information on this and I can't find anything useful.  I'm currenty
building a realtime visualization system for a Space Shuttle landing
simulator at NASA.  I'm using a small network of 5 Pentium 4 computers
running RedHat's 2.4.20-31.9 kernel.  I'm easily running 60 frames/second on
my systems, but I'm having a problem because the kscand/Normal thread comes
in every 25 seconds and causes me to drop a frame (very annoying).  I've
looked into the kernel source and found where the kscand threads are
spawned.  I also see where the 25 second period is coming from.  What I'm
wondering is what would happen if I disabled the kscand/Normal thread?  I've
got plenty of memory, and my process is the only thing running on the
system.  Would I eventually see problems, or would I be OK since I'm not
running low on memory?  What if I modified the kernel to allow me to
temporarily disable the thread while my application is running (using a
/proc file or something similar)?  Sorry if this is a bad question, but I
figure the people on this list are the best source of info.

Please CC: me directly since I'm not subscribed to the list. 

Thanks for any help or suggestions,
Corbin
corbin.l.holtz@jsc.nasa.gov

Re: Disable kscand/Normal?

[Con Kolivas]

HOLTZ, CORBIN L. (JSC-ER) (LM) writes:

> Hello,
> 
> Sorry to post to the list without being subscribed, but I've searched the
> web for information on this and I can't find anything useful.  I'm currenty
> building a realtime visualization system for a Space Shuttle landing
> simulator at NASA.  I'm using a small network of 5 Pentium 4 computers
> running RedHat's 2.4.20-31.9 kernel.  I'm easily running 60 frames/second on
> my systems, but I'm having a problem because the kscand/Normal thread comes
> in every 25 seconds and causes me to drop a frame (very annoying).  I've
> looked into the kernel source and found where the kscand threads are
> spawned.  I also see where the 25 second period is coming from.  What I'm
> wondering is what would happen if I disabled the kscand/Normal thread?  I've
> got plenty of memory, and my process is the only thing running on the
> system.  Would I eventually see problems, or would I be OK since I'm not
> running low on memory?  What if I modified the kernel to allow me to
> temporarily disable the thread while my application is running (using a
> /proc file or something similar)?  Sorry if this is a bad question, but I
> figure the people on this list are the best source of info.
> 
> Please CC: me directly since I'm not subscribed to the list. 

That vendor kernel that you're running has the rmap vm which has the kscand 
daemon. If you build a newer vanilla kernel it will not have the kscand 
daemon. Alternatively, you can manually nice the the kscand daemon as root 
to a lower value (not sure what it is by default) say nice +19. You can also 
rebuild your vendor's kernel and edit the code to set the nice level 
on spawning the daemon (obviously this requires more knowledge than the 
previous options).

Cheers,
Con

Re: Disable kscand/Normal?

[Lee Revell]

On Wed, 2004-08-25 at 23:54, HOLTZ, CORBIN L. (JSC-ER) (LM) wrote:
> I'm currenty
> building a realtime visualization system for a Space Shuttle landing
> simulator at NASA.  I'm using a small network of 5 Pentium 4 computers
> running RedHat's 2.4.20-31.9 kernel.  I'm easily running 60 frames/second on
> my systems, but I'm having a problem because the kscand/Normal thread comes
> in every 25 seconds and causes me to drop a frame (very annoying).  I've
> looked into the kernel source and found where the kscand threads are
> spawned.  I also see where the 25 second period is coming from.  What I'm
> wondering is what would happen if I disabled the kscand/Normal thread?  I've
> got plenty of memory, and my process is the only thing running on the
> system.  Would I eventually see problems, or would I be OK since I'm not
> running low on memory?  What if I modified the kernel to allow me to
> temporarily disable the thread while my application is running (using a
> /proc file or something similar)?

You should also look into Ingo Molnar's voluntary preemption patches for
a more general way to do soft and even hard realtime with Linux.  Con's
suggestion will probably solve the kscand problem but the voluntary
preemption patches provide a more general way to deal with real time
constraints.  Check the LKML archives for the past few months, there has
been a lot of work in this area lately.

Lee