Re: [PATCH 11/32] nohz/cpuset: Don't turn off the tick if rcu needsit

[Steven Rostedt]

On Tue, 2012-03-27 at 20:19 -0500, Christoph Lameter wrote:
> On Tue, 27 Mar 2012, Steven Rostedt wrote:
> 
> > > Obviously compiling a kernel with preemptiong introduces additional
> > > overhead to guarantee more deterministic behavior. Additional overhead
> > > increases latencies generated by the OS in general. Compile a kernel
> > > without preemption and it will run faster and thus have lower latencies.
> >
> > I call that "lower overhead".
> 
> Good marketing but it does not change the facts.

I see we are mixing the paint for the bike shed.

> 
> > "Latency is a measure of time delay experienced in a system, the precise
> > definition of which depends on the system and the time being measured.
> > Latencies may have different meaning in different contexts."
> >
> > That last sentence is key. So lets avoid the term "latency" as it
> > obviously has a different meaning to the both of us.
> >
> > Instead, lets use "determinism" (what we call latency in the realtime
> > world) and "overhead" (what you seem to see as latency caused by the
> > kernel).
> 
> I sure wish you would be using the term determinism instead of "latency".
> 
> Overhead causes latency and the definition that you quoted is what I am
> talking about. Latencies are the delays in processing experienced by the
> application through the speed of system calls and by interruptions  of
> a user space process by the kernel for various reasons.

I could also argue that a non-preempt kernel has a large latency as
well. Although it may have good through put for one task, another task
may suffer from a large latency waiting for a lower priority task to get
out of a system call.

You say tomAYto I say tomAHto.

Read the article: http://en.wikipedia.org/wiki/Latency_%28engineering%29

Especially the section about: Computer hardware and operating system latency

You'll see that it describes latency much closer to my definition than
yours.

Heck, google "operating system latency" and you'll see a lot of talk
about reaction times and how fast the hardware can do its job. I don't
see anything about the time a system call takes.

-- Steve



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/