MFC of UDP socket performance test

[garcol at postino.it]

On Thu, 15 Mar 2007, Marcelo Gardini do Amaral wrote:

> I repeated that performance test done with bind [1] using now Robert's MFC 
> [2]. Another tweak that I was supposed to do was to use libthr instead of 
> libpthread (via libmap.conf) and build bind with threads option. In the 

new 
> test I did this.
>
> I used the same methodology, same zone file and same server hardware that 

I 
> used in [3].



Hi,
   I measured the response performance as "Queries sent" with the queryperf 
program for 30s and for 6 times. 
queryperf asked various host names randomly generated.
queryperf and named run on the same server.


CPU: 2x Pentium III/Pentium III Xeon/Celeron (601.37-MHz 686-class CPU)
named_freebsd:9.3.3
named_ports:9.3.4
named_ports_th: +thread  (-n #cpus)
named_ports2:9.4.0
named_ports2_th: +thread  (-n #cpus)


6.2-RELEASE-p2
named  

named_ports  named_ports_th
             th1 
             th2 

6.2-RELEASE-p2 SMP
named_freebsd  
72367
named_ports  named_ports_th
74287        -n1 50228
             -n2 43579

6.2-RELEASE-p2 SMP+PATCH(sosend_copyin.diff)
named_ports  named_ports_th    named_ports2_th
74260        -n1 49544
             -n2 43382		

6.2-RELEASE-p2 SMP+PATCH2(sosend_dgram.diff)
named_ports  named_ports_th    named_ports2_th
74004        -n1 49358
             -n2 53232

I make this change:
/etc/libmap.conf
[named9-th]
libpthread.so.2 libthr.so.2
libpthread.so libthr.so

6.2-RELEASE-p2 SMP
named_ports_th
-n1 60228
-n2 96579  * (but the idle CPU is 0%)

6.2-RELEASE-p2 SMP+PATCH(sosend_copyin.diff)
named_ports_th
-n1 60650
-n2 96316  * (but the idle CPU is 0%)

6.2-RELEASE-p2 SMP+PATCH2(sosend_dgram.diff)
named_ports_th
-n1 60462
-n2 95976  * (but the idle CPU is 0%)


I've seen with libthr the performance increase significatly according with 
Marcelo.

Regards
Alessandro