What to do with these PCs?

Yeah, acpi was the first one that I noticed as well. All those processes are the defaults of Debian Wheezy 7.2 with XFCE desktop. I installed from CD, setup wifi and then ran 'apt-get update' and 'apt-get upgrade'. Mousepad is the text editor that has come with xfce....no, I've never used it before either...

The motherboards were the other way round - the ASUS/HP board had the best latency (<8K with the Ubuntu 8.04 LiveCD). The MSI K8 board had reasonable results but was randomly freezing with Ubuntu 10.04. I haven't tried any other OS on it, so don't know if it was a hardware issue or a software/module conflict. Only one way to find out...

Hi to all

I run into a silly idea.

If I understand right, latency comes as there are (modern) hw components that generate signals we can not block.

Why not fullfill ?

I gues what happens is that by such component an interupt is generated, but no driver responses, system goes into some kind of wait until a superiour timeout clears the request.

My idea:
One (kernel?) module that reads at startup those "not to react" interupts from a file.
As soon as such an interupt is generated it is answered with null operation.

Ok, I'm not knowing linux in detail...

Mike

Hi Mike

Unfortunately that is not how it works. Might be nice if it was.

The hardware interrupt issues with latency largely occur, not because there is no receiver for the interrupt, but because they completely bypass the realtime kernel task schedualing and interrupt realtime processes.

This is at the heart of the SMI (System Maintainence Interrupt) issue

A typical example of this is temperature sensing fan control on some machines.
The polling of kernel temperature and setting fan speed to suit can be very disruptive to latency, because it is a high priority low level BIOS function with no regard to whatever OS is sat above it.

The answer in that case is usually to set the fan to full speed all the time, if that option exists, so that no temperature sensing is required.

regards

Hi Arceye

But.

I like your explanation. Just link inbetween before SMI interupt chain.. .Than you can ignore the unneeded and send the rest to the normal interupt handler. This should prolongate only by a few thousands of native cpu cycles..
I don't know how to do that. Too many years when I started by pure asembler... Lateron RSX .. And quite some 68000 industrial realtime systems..
If you go down to the bottom in earlier days there have allways been some interupt catchup registers.
Don't know, but think linux is similar..

Mike