Netra T1 105 (sparc64) optimization

On 11/30/06, Andrew Belashov <belorel.ru> wrote:
>
> Hello, All!
>
> Miles Nordin wrote:
> >>>>>> "av" == Alexandre
Vieira <nullptgmail.com> writes:
> >
> >     av> kernel/tunables recommendations,
> >
> > I suggest device polling.  This will increase the
pps you can forward,
> > and optimistically cause the machine to drop
packets rather than go
> > into livelock if the pps exceeds its ability. 
IMHO it should be used
> > on any router.  and IIRC it does work with the hme
driver.
>
> I use device polling(4) patch for hme(4):
> <
> http:/
/docs.freebsd.org/cgi/getmsg.cgi?fetch=17147+0+archive/2006/
freebsd-sparc64/20060423.freebsd-sparc64
> >
>
> --
> With Best Regards,
> Andrew Belashov.
>
>
Hi,

Thanks for the tips!

I've added the patch for hme polling and I'm recompiling the
kernel. I've
striped the kernel a lot and I'm compiling with -pipe -O2
-fno-strict-aliasing.

AFAIK there isn't any CPUTYPE available for this cpu.

TIA
-- 
Alexandre Vieira - nullptgmail.com
_______________________________________________
freebsd-questionsfreebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-que
stions
To unsubscribe, send any mail to
"freebsd-questions-unsubscribefreebsd.org"

Hi,

the Netra comes with a "360- or 440- MHz
UltraSPARC-IIi, single
processor, Superscalar SPARC Version 9" CPU, so you can
use
"--mcpu=ultrasparc --mtune=ultrasparc" with gcc.
This is for gcc v3.x,
gcc v2.x might require --march instead of --mtune.

HTH
Christian
_______________________________________________
freebsd-questionsfreebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-que
stions
To unsubscribe, send any mail to
"freebsd-questions-unsubscribefreebsd.org"

On 11/30/06, Christian Walther <cptsalekgmail.com> wrote:
>
> Hi,
>
> the Netra comes with a "360- or 440- MHz
UltraSPARC-IIi, single
> processor, Superscalar SPARC Version 9" CPU, so
you can use
> "--mcpu=ultrasparc --mtune=ultrasparc" with
gcc. This is for gcc v3.x,
> gcc v2.x might require --march instead of --mtune.
>
> HTH
> Christian
>

Thanks for the tip, I will use this gcc options.

Miles/Andrew polling increased dramaticly the latency for
all traffic that
passes trough the bridge and reduced the transmission rate
also.

with polling:

rtt min/avg/max/mdev = 1.472/2.005/2.651/0.393 ms

without polling:

rtt min/avg/max/mdev = 0.642/0.735/1.111/0.134 ms

Thanks in advance
-- 
Alexandre Vieira - nullptgmail.com
_______________________________________________
freebsd-questionsfreebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-que
stions
To unsubscribe, send any mail to
"freebsd-questions-unsubscribefreebsd.org"