DNS query performance

[Marcelo Gardini do Amaral]

> >Are you able to boot a 7.x kernel on this box?  An as yet un-MFC'd 
> >optimization to the UDP send path is present in the 7.x kernel, suggested 
> >by ISC, which significantly improves threaded BIND9 performance.  I've not 
> >benchmarked unthreaded BIND9 with the change.  If you want to test 
> >specifically the before/after case for that change, you can find the 
> >reference to sosend_dgram in src/sys/netinet/udp_usrreq.c and swap it to 
> >sosend, which restores the old behavior.

I booted 7.x kernel UP and SMP on my blade. When I swaped both of
them to sosend I got a panic to any DNS query:

# dig @localhost test1.foo.bar                                    
panic: sosend: protocol calls sosend                                           
KDB: enter: panic                                    
[thread pid 671 tid 100053 ]                         
Stopped at      kdb_enter+0x2b: nop                  
db>                                                

db> bt                                               
Tracing pid 671 tid 100053 td 0xc4a52bd0             
kdb_enter(c091d7b6) at kdb_enter+0x2b                
panic(c0925143,e502dc10,c06e6511,c4f9c14c,c4a66370,...) at panic+0xbb
sosend(c4f9c14c,c4a66370,e502dbe4,0,0,0,c4a52bd0) at sosend+0x1f
kern_sendit(c4a52bd0,15,e502dc5c,0,0,0) at kern_sendit+0x101
sendit(c4a52bd0,15,e502dc5c,0,c4a66150,...) at sendit+0x87
sendmsg(c4a52bd0,e502dd04) at sendmsg+0x53           
syscall(3b,3b,3b,1,0,...) at syscall+0x256               
Xint0x80_syscall() at Xint0x80_syscall+0x1f          
--- syscall (28, FreeBSD ELF32, sendmsg), eip = 0x28353d6f, esp =
0xbfbfe8ec, eb
p = 0xbfbfea68 ---

db> c                                                                           
Uptime: 1m44s                                                                   
Physical memory: 2039 MB                                                        
Dumping 99 MB: 84 68 52 36 20 4                                                 
Dump complete                                                                   
Automatic reboot in 15 seconds - press a key on the console to abort            


> The other common optimization advice that you may already have received is 
> to check which time counter FreeBSD has selected.  Right now, 6.x/7.x err 
> on the side of accurate over fast.  There's been quite a bit of debate 
> about this approach, and it's useful to investigate the issue.  You can 
> view and set the current choice by looking at the sysctl 
> kern.timecounter.hardware, and you can see the choices on your hardware by 
> looking at kern.timecounter.choice. Typically, TSC is the fastest, but may 
> suffer from drift as the CPU changes speed (as a result of temperature, 
> power saving, etc).  Set it to TSC if it's not already TSC, and see what 
> the effect is.  As many event libraries read time stamps frequently to set 
> up sleeping in user space, it can have a substantial performance impact.

With the 7.x kernel and no changes in src/sys/netinet/udp_usrreq.c I
tried different timecounters and I couldn't see any performance
difference. Here we see the results for bind 9.3.2, same zone file and
queries:


Kernel UP

Timecounter	queries/s
-----------	---------

ACPI-safe	16200

TSC		16584

i8254		16319



Kernel SMP

Timecounter     queries/s
-----------     ---------

ACPI-safe       15323

TSC             15930

i8254           14155



Any other tip?


-- 
Att.,

Marcelo Gardini