[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: 10k pps



On Fri, Apr 07, 2006 at 12:17:58AM +0200, Per-Olov Sjvholm wrote:
> On Thursday 06 April 2006 23.08, Claudio Jeker wrote:
> > On Thu, Apr 06, 2006 at 11:47:16PM +0300, Claudiu Pruna wrote:
> > > 	Hi there list,
> > >
> > > 	I got to a situation at work where I have an OpenBSD 3.9 amd64 router
> > > acting as bgp and ospf router, and it has to coupe with 100Mbps and
> > > approx 15.000 packets per second, but it can't at about 10k pps, I have
> > > like 70% cpu utilisation on iterrupt, and all the traffic becomes an
> > > extreme sport, it is an Intel P4 3GHz em64 with 512MB of ram and 2 Intel
> > > Pro100 (fxp) network cards.
> > >
> > > 	Any ideea if/how can I "jump" over the 10k barrier ?
> > >
> > >
> > >
> > > P.S.: Claudio thanks for the advice about 3.9 bgpd version and additive
> > > communities, I works smooth.
> > >
> > > Thanks for any sugestion or advice.
> >
> > Switch to i386. amd64 has some interrupt problems, the amd64 I tested once
> > maxed at 80kpps but did 450kpps in i386 mode.
> 
> Hi Claudio
> 
> What cpu, network cards and pf ruleset size did you use during the test when 
> the server handled 450kpps ? 
> 

CPU (actually two CPUs on the board):
cpu0 at mainbus0: (uniprocessor)
cpu0: AMD Engineering Sample, 2592.68 MHz
cpu0: FPU,VME,DE,PSE,TSC,MSR,PAE,MCE,CX8,APIC,SEP,MTRR,PGE,MCA,CMOV,PAT,PSE36,CF
LUSH,MMX,FXSR,SSE,SSE2,SSE3,NXE,MMXX,FFXSR,LONG,3DNOW2,3DNOW
cpu0: 64KB 64b/line 2-way I-cache, 64KB 64b/line 2-way D-cache, 1MB 64b/line 16-way L2 cache
cpu0: ITLB 32 4KB entries fully associative, 8 4MB entries fully associative
cpu0: DTLB 32 4KB entries fully associative, 8 4MB entries fully associative

Network cards:
bge0 at pci2 dev 9 function 0 "Broadcom BCM5704C" rev 0x03, BCM5704 A3 (0x2003):
 irq 10, address 00:e0:81:27:e0:a9
brgphy0 at bge0 phy 1: BCM5704 10/100/1000baseT PHY, rev. 0
bge1 at pci2 dev 9 function 1 "Broadcom BCM5704C" rev 0x03, BCM5704 A3 (0x2003):
 irq 5, address 00:e0:81:27:e0:aa
brgphy1 at bge1 phy 1: BCM5704 10/100/1000baseT PHY, rev. 0

PF was disabled (enabling PF with 10 or 20 rules (no states) resulted in a
20-30% drop)

At the time we measured it em(4) was slower (300-350kpps) but fixes went in
to remove the bottlenecks in the em(4) driver.
-- 
:wq Claudio