cvs commit: src/sys/dev/hatm if_hatm.c if_hatm_intr.c if_hatmvar.h

[Harti Brandt]

On Wed, 29 Oct 2003, John Baldwin wrote:

JB>
JB>On 29-Oct-2003 Hartmut Brandt wrote:
JB>> harti       2003/10/29 04:59:44 PST
JB>>
JB>>   FreeBSD src repository
JB>>
JB>>   Modified files:
JB>>     sys/dev/hatm         if_hatm.c if_hatm_intr.c if_hatmvar.h
JB>>   Log:
JB>>   Get rid of the mutexes for the exernal buffer free lists. Use
JB>>   atomic instructions instead. Remove the stuff used to track
JB>>   whether an external mbuf travels through the system. This is
JB>>   temporary only and will come back soon.
JB>
JB>Do the atomic operations buy you significant performance?

Yes. I'm trying to receive and send around 150kpkt/sec. Up to now it
stopped at 120kpkts because the (2-CPU) machine was 50% in interrupt
state. Now I'm up to 180kptks with around 45%. I'm not sure how much of
this is due to getting rid of the mutexes, but I may have to try it with
the mutexes because I got another mail that says, that the algorithm may
not be safe. I'll take a look.

harti
-- 
harti brandt,
http://www.fokus.fraunhofer.de/research/cc/cats/employees/hartmut.brandt/private
brandt at fokus.fraunhofer.de, harti at freebsd.org