Re: [RFC PATCH RT V3] rwsem: The return of multi-reader PI rwsems

[Steven Rostedt <rostedt@goodmis.org>]

On Fri, 11 Apr 2014 14:47:49 +0200
Carsten Emde <C.Emde@osadl.org> wrote:

> Hi Steven,
> 
> > [..] I added Carsten to the Cc, so I'll post the entire change log
> > of v1 here again.
> I've been listening and testing boxes all the time ...
> 

I figured but still wanted to add you to the Cc.

> > [..] If you have any benchmark on large machines I would be very
> > happy if you could test this patch against the unpatched version of
> > -rt.
> Three machines
> - an X32 x86_64 (AMD Opteron 6272 @2100 MHz) at rack #1/slot #1,
> - an X4x2 x86_64 (Intel i7-2600K @3400 MHz) at rack #4/slot #6, and
> - an X4 ARM (i.MX6 Quad @996 MHz) at rack #8/slot #7
> are running a v3-patched 3.12.15-rt25 kernel now. I'll equip more
> machines later.
> 
> What I can say so far is:
> - No evidence for any regression, no crashes

That's good to hear.

> - Performance certainly at least as good as unpatched, probably better

That's even better.

> 
> I'll do more tests and come back with more precise performance
> comparison data.

Do you also have any threaded tests? That is, something like a java
benchmark that kicks off lots of threads. That's where the performance
should show up the most. Clark's whack_mmap_sem test is specific to
this, and has shown an 10x increase in performance with my patch. But
that's a micro benchmark. A better test would be a real java
application.

Thanks,

-- Steve