Date: Mon, 31 Aug 2009 18:35:47 -0700 From: Kip Macy <kmacy@freebsd.org> To: fabient@freebsd.org Cc: freebsd-current@freebsd.org Subject: Re: Forwarding benchmark Message-ID: <3c1674c90908311835i7f70e6d8mbfa8ce619e7ff911@mail.gmail.com> In-Reply-To: <D0E9834C-59EB-48B2-A804-6947DA75C7D6@netasq.com> References: <D0E9834C-59EB-48B2-A804-6947DA75C7D6@netasq.com>
next in thread | previous in thread | raw e-mail | index | archive | help
We're not going to see much more 700kpps on forwarding workloads until we do something about the rtentry locking. I had some interesting ideas I was exploring, but I don't have the luxury of side projects right now. em(9)'s transmit performance has been substantially improved in 8 by using a buf_ring instead of IFQ so I assume that you're entirely gated by rx performance. Jeff did some work in that area to reduce the per-packet overhead of dequeue and to do some NAPI-like opportunistic polling using a variant of the taskqueue API. It won't give you any idea about latency breakdown, but it would be useful for a general time breakdown to look at unhalted core cycles in PMC. Good Luck, Kip On Fri, Aug 21, 2009 at 02:25, Fabien Thomas<fabien.thomas@netasq.com> wrot= e: > =A0 =A0 =A0 =A0Hi all, > > Just a quick benchmark on 8.0 Beta2+ (18/08) show no regression vs 7.2. > > Result in FPS for 64bytes frame using Breakingpoint Elite > > Breakingpoint P1 =3D=3D=3D DUT =3D=3D=3D Breakingpoint P2 > > Stream1 : P1 -> P2 > Stream2: P2 -> P1 > > GENERIC kernel + netisr.direct > > 4.11 : 236 (with 1 stream down for unknown reason) > 6.3 =A0 : 248 > 7.2 =A0 : 350 > 8.0b : 352 > > POLLING kernel + netisr.direct > > 4.11 : 526 > 6.3 =A0 : 246 > 7.2 =A0 : 230 > 8.0b : 330 > > Note that the perf grow a little bit from version to version but 4.11 wit= h > polling is always a lot better. > > There is a lot a more in depth testing to do (HW flow tag, 10gb, lot of > interface, latency ...) but it give a rough idea of the perf in the > forwarding area. > > Regards, > Fabien > > dmesg: > > CPU: Intel(R) Pentium(R) D CPU 2.80GHz (2793.02-MHz 686-class CPU) > =A0Origin =3D "GenuineIntel" =A0Id =3D 0xf47 =A0Stepping =3D 7 > =A0Features=3D0xbfebfbff<FPU,VME,DE,PSE,TSC,MSR,PAE,MCE,CX8,APIC,SEP,MTRR= ,PGE,MCA,CMOV,PAT,PSE36,CLFLUSH,DTS,ACPI,MMX,FXSR,SSE,SSE2,SS,HTT,TM,PBE> > =A0Features2=3D0x641d<SSE3,DTES64,MON,DS_CPL,CNXT-ID,CX16,xTPR> > =A0AMD Features=3D0x20100000<NX,LM> > =A0AMD Features2=3D0x1<LAHF> > =A0TSC: P-state invariant > real memory =A0=3D 1073741824 (1024 MB) > avail memory =3D 1035210752 (987 MB) > ACPI APIC Table: <PTLTD =A0 =A0 =A0 =A0 =A0APIC =A0> > FreeBSD/SMP: Multiprocessor System Detected: 2 CPUs > FreeBSD/SMP: 1 package(s) x 2 core(s) > =A0cpu0 (BSP): APIC ID: =A00 > =A0cpu1 (AP): APIC ID: =A01 > ... > em8: <Intel(R) PRO/1000 Network Connection 6.9.14> port 0x7000-0x701f mem > 0xed700000-0xed71ffff irq 18 at device 0.0 on pci6 > em8: Using MSI interrupt > em8: [FILTER] > em8: Ethernet address: 00:30:48:5c:40:82 > pcib7: <ACPI PCI-PCI bridge> irq 19 at device 28.3 on pci0 > pci8: <ACPI PCI bus> on pcib7 > em9: <Intel(R) PRO/1000 Network Connection 6.9.14> port 0x8000-0x801f mem > 0xed800000-0xed81ffff irq 19 at device 0.0 on pci8 > em9: Using MSI interrupt > em9: [FILTER] > em9: Ethernet address: 00:30:48:5c:40:83 > > _______________________________________________ > freebsd-current@freebsd.org mailing list > http://lists.freebsd.org/mailman/listinfo/freebsd-current > To unsubscribe, send any mail to "freebsd-current-unsubscribe@freebsd.org= " > --=20 When harsh accusations depart too far from the truth, they leave bitter consequences. --Tacitus
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3c1674c90908311835i7f70e6d8mbfa8ce619e7ff911>