Early versions of the ixgbe driver unperformed when compared to the igb driver for small packet performance. I actually gave a presentation on this a couple of years ago: http://www.linuxplumbersconf.org/2011/ocw/system/presentations/423/original/optimizing-ixgbe-performance.pdf
My advice would be to update to more recent versions of the igb and ixgbe driver for your comparison. That should help to improve the performance seen for both drivers. Thanks, Alex On 09/13/2013 08:50 AM, Skidmore, Donald C wrote: > Hey Abhinay, > > I would be interested in seeing your statics on the ports before/after a test > (ethtool -s ethX). Likewise what your interrupt distribution looks like > before/after (cat /proc/interrupts | grep ethX). lscpi -vvv could be useful > as well. > > Thanks, > -Don Skidmore <[email protected]> > >> -----Original Message----- >> From: arun abhinay [mailto:[email protected]] >> Sent: Friday, September 13, 2013 4:13 AM >> To: zhuyj >> Cc: [email protected] >> Subject: Re: [E1000-devel] Traffic forwarding rate of 10G interfaces is lower >> than that of 1G interfaces >> >> Hi zhuyi, >> >> Thanks for your reply. UDP traffic is used for my testing. Only one core was >> being used for the trasmission of packets on both 1G and 10G interfaces. >> >> Can you let us know your opinion and inputs which would help us fix this >> issue.. >> >> Thanks >> Abhinay >> >> On Fri, Sep 13, 2013 at 10:01 AM, zhuyj <[email protected]> wrote: >> >>> With tcp or raw IP? >>> >>> Best Regards! >>> Zhu Yanjun >>> >>> >>> On 09/13/2013 03:17 PM, arun abhinay wrote: >>> >>> Hi, >>> >>> Please redirect me if this is not the correct forum for my question. >>> >>> I am trying to check the thourgh put rate on 10G and 1G interfaces >>> using traffic generator with configuration as mentioned below on a >>> x86_64 machine. >>> >>> 1G Iface setup >>> Traffic Generator Port1 ----> eth0 <x86_64 machine> eth1 ---> Traffic >>> Generator Port2 >>> >>> 10G Iface setup >>> Traffic Generator Port3 ----> eth2 <x86_64 machine> eth3 ---> Traffic >>> Generator Port4 >>> >>> Only 1G ifaces or 10G ifaces are tested at any time and Traffic is >>> full duplex. >>> Traffic generator port1 and port3 has 100 IP addresses configured, >>> Traffic generator ports2 and 4 have only 1 IP address. This means that >>> traffic is send from 100 source addresses to 1 destination address and >>> (because full duplex traffic) from 1 source address to 100 destination >>> addresses. Result is 200 different source-destination pairs (I.e 200 >>> "flows"). >>> >>> When comparing maximum packets per second forwarding rate of 1G >>> interfaces >>> (eth0 and eth1) to the maximum forwarding rate of 10G interfaces (eth2 >>> and eth3), the 10G interfaces is slower with small packets (74 bytes). >>> With small IP packets (74Byte) the limiting factor is the CPU processing >> power. >>> At the maximum pps rate , the CPU core is loaded 100% in both 1G and >>> 10G ifaces case. Please find the attached excel sheets which show >>> thorough put rates on 1G and 10G interfaces with various packet sizes. >>> With 74Byte packet and 1G ifaces the bit rate is about 584Mbps whereas >>> 10G ifaces and 74Byte packet the max is about 278Mbps. >>> >>> Can you please check and let us know why the thoughput rate of 10G >>> interfaces is less than throughput rate of 1G interfaces with smaller >>> packet frames. Is the ixgbe driver less efficient than igb driver with >>> small packets. Please let me know if any futher information is required. >>> The ethernet driver details and settings are as mentioned below Driver >>> information >>> ------------------------- >>> 1G ethernet ports eth0 and eth1 >>> # ethtool -i eth0 >>> driver: igb >>> version: 3.1.16 >>> firmware-version: 0.9-1 >>> bus-info: 0000:01:00.1 >>> 10G ethernet ports eth0 and eth1 >>> # ethtool -i eth2 >>> driver: ixgbe >>> version: 3.1.17-NAPI >>> firmware-version: 4.2-1 >>> bus-info: 0000:02:00.0 >>> >>> Setting are same on 1G and 10G interfaces >>> --------------------------------------------------------- >>> 1G iface: >>> # ethtool -k eth0 >>> Offload parameters for eth0: >>> rx-checksumming: on >>> tx-checksumming: on >>> scatter-gather: on >>> tcp-segmentation-offload: on >>> udp-fragmentation-offload: off >>> generic-segmentation-offload: on >>> generic-receive-offload: on >>> large-receive-offload: off >>> >>> 10G iface: >>> # ethtool -k eth2 >>> Offload parameters for eth2: >>> rx-checksumming: on >>> tx-checksumming: on >>> scatter-gather: on >>> tcp-segmentation-offload: on >>> udp-fragmentation-offload: off >>> generic-segmentation-offload: on >>> generic-receive-offload: on >>> large-receive-offload: off >>> ------------------------------------------------------------------------------ How ServiceNow helps IT people transform IT departments: 1. Consolidate legacy IT systems to a single system of record for IT 2. Standardize and globalize service processes across IT 3. Implement zero-touch automation to replace manual, redundant tasks http://pubads.g.doubleclick.net/gampad/clk?id=51271111&iu=/4140/ostg.clktrk _______________________________________________ E1000-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/e1000-devel To learn more about Intel® Ethernet, visit http://communities.intel.com/community/wired
