Hi Mirko, Thanks for your report and all the great information you gathered.
I have a couple additional questions - What were you running to do this test? (i.e. netperf, pktgen?) - What throughput were you seeing? - Are both ports plugged into x8 PCIe slots - Did you look at the dmesg for anything relative? - Do you see this issue only with UDP? What I'm wondering is if this NIC was receiving faster than it could process them. Thanks, -Don Skidmore <[email protected]> >-----Original Message----- >From: Mirko Corosu [mailto:[email protected]] >Sent: Friday, February 19, 2010 5:30 AM >To: [email protected] >Subject: [E1000-devel] Intel 2598EB 10-Gigabit AT dropped rx packet > >Dear all, > >I'm new to this list, if this isn't the right place to post this issue, >please let me know. > >I am testing two Intel 82598EB 10-Gigabit AT network card mounted on two >Dell R710 server with Red Hat Enterprise Linux 5.4 installed. The two >NICs are directly connected each other (no switch in between). I am >experiencing a problem on the receive side of the connection: I'm sending >an UDP stream after having disabled flow control on each card, and about >the 60% of the transmitted packets are dropped with a large number of >rx_missed_errors. This is what ifconfig and ethtool -S show: > >[r...@grids2]# ethtool -S eth0 >NIC statistics: > rx_packets: 1761794 > tx_packets: 32 > rx_bytes: 14748667944 > tx_bytes: 5746 > lsc_int: 2 > tx_busy: 0 > non_eop_descs: 6510828 > rx_errors: 0 > tx_errors: 0 > rx_dropped: 0 > tx_dropped: 0 > multicast: 2 > broadcast: 2 > rx_no_buffer_count: 0 > collisions: 0 > rx_over_errors: 0 > rx_crc_errors: 0 > rx_frame_errors: 0 > rx_fifo_errors: 0 > rx_missed_errors: 1148382 > tx_aborted_errors: 0 > tx_carrier_errors: 0 > tx_fifo_errors: 0 > tx_heartbeat_errors: 0 > tx_timeout_count: 0 > tx_restart_queue: 0 > rx_long_length_errors: 0 > rx_short_length_errors: 0 > tx_tcp4_seg_ctxt: 0 > tx_tcp6_seg_ctxt: 0 > tx_flow_control_xon: 0 > rx_flow_control_xon: 0 > tx_flow_control_xoff: 0 > rx_flow_control_xoff: 0 > rx_csum_offload_good: 28 > rx_csum_offload_errors: 0 > tx_csum_offload_ctxt: 11 > low_latency_interrupt: 0 > alloc_rx_page_failed: 0 > alloc_rx_buff_failed: 0 > lro_aggregated: 8 > lro_flushed: 8 > lro_recycled: 0 > rx_no_dma_resources: 0 > hw_rsc_count: 0 > rx_flm: 0 > fdir_match: 0 > fdir_miss: 0 > tx_queue_0_packets: 32 > tx_queue_0_bytes: 5746 > rx_queue_0_packets: 2 > rx_queue_0_bytes: 120 > rx_queue_1_packets: 2 > rx_queue_1_bytes: 120 > rx_queue_2_packets: 1761762 > rx_queue_2_bytes: 14748664800 > rx_queue_3_packets: 0 > rx_queue_3_bytes: 0 > rx_queue_4_packets: 0 > rx_queue_4_bytes: 0 > rx_queue_5_packets: 14 > rx_queue_5_bytes: 1452 > rx_queue_6_packets: 7 > rx_queue_6_bytes: 726 > rx_queue_7_packets: 7 > rx_queue_7_bytes: 726 > >[r...@grids2]# ifconfig eth0 > >eth0 Link encap:Ethernet HWaddr 00:1B:21:4B:C8:BF > inet addr:10.100.100.2 Bcast:10.100.100.255 Mask:255.255.255.0 > inet6 addr: fe80::21b:21ff:fe4b:c8bf/64 Scope:Link > UP BROADCAST RUNNING MULTICAST MTU:9000 Metric:1 > RX packets:1761795 errors:0 dropped:1148382 overruns:0 frame:0 > TX packets:39 errors:0 dropped:0 overruns:0 carrier:0 > collisions:0 txqueuelen:1000 > RX bytes:14748668004 (13.7 GiB) TX bytes:9332 (9.1 KiB) > >Each interface has the MTU set to 9000, the driver is the one provided >by Dell support (ixgbe 2.0.44.14-NAPI) with Rx multiqueue enabled and >InterruptThrottleRate set to 8000 (but I tried 16000 and 32000 too). >Also I tried to set FdirPballoc parameter to 0, 1 and 2 with no success. >Single CPU load is no more than 10% with 8000 interrupt/s: > >[r...@grids2 ~]# mpstat -P ALL 2 > >05:16:43 PM CPU %user %nice %sys %iowait %irq %soft %steal >%idle intr/s >05:16:45 PM all 0.00 0.00 0.06 0.00 0.06 1.19 0.00 > 98.69 9018.00 >05:16:45 PM 0 0.00 0.00 0.00 0.00 0.00 0.00 0.00 > 100.00 1002.50 >05:16:45 PM 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 > 100.00 1.50 >05:16:45 PM 2 0.00 0.00 0.50 0.00 0.50 9.45 0.00 > 89.55 8007.00 >05:16:45 PM 3 0.00 0.00 0.00 0.00 0.00 0.00 0.00 > 100.00 1.00 >05:16:45 PM 4 0.00 0.00 0.00 0.00 0.00 0.00 0.00 > 100.00 1.00 >05:16:45 PM 5 0.00 0.00 0.00 0.00 0.00 0.00 0.00 > 100.00 3.00 >05:16:45 PM 6 0.00 0.00 0.00 0.00 0.00 0.00 0.00 > 100.00 1.00 >05:16:45 PM 7 0.00 0.00 0.00 0.00 0.00 0.00 0.00 > 100.00 1.00 > >I tried to configure smp affinity to assign the IRQ of each >rx-queue to a different CPU, with no effect on the amount of dropped >packets. I also verified that if I connect the NIC on a 10GE switch and >send more than one stream from different servers, the card uses >different rx-queues but the problem still exists. > >I modified the /etc/sysctl.conf file parameters adding: > >net.core.wmem_max = 16777216 >net.core.wmem_default = 8388608 >net.core.rmem_max = 16777216 >net.core.rmem_default = 8388608 >net.ipv4.tcp_rmem = 4096 262144 16777216 >net.ipv4.tcp_wmem = 4096 262144 16777216 >net.ipv4.tcp_mem = 4096 8388608 16777216 >net.core.optmem_max = 524288 >net.core.netdev_max_backlog = 200000 > > > > >Some other details: > >- CPU: dual quad core Intel L5520 @ 2.27GHz (Hyper-Threading disabled) >- RAM: 24 GB (12 X 2GB) >- Slot PCI: PCIe gen.2 x8 >- Kernel linux: 2.6.18-128.1.1.el5 and 2.6.18-164.2.1.el5 > >Any suggestions? >Thank you very much in advance. > >Mirko Corosu > >-- > >-------------------------------------------------------------------------- >Mirko Corosu >Network and system administrator >Computing Center >Istituto Nazionale Fisica Nucleare >Via Dodecaneso 33 >16146 Genova, Italy >http://www.ge.infn.it >-------------------------------------------------------------------------- > > > >--------------------------------------------------------------------------- >--- >Download Intel® Parallel Studio Eval >Try the new software tools for yourself. Speed compiling, find bugs >proactively, and fine-tune applications for parallel performance. >See why Intel Parallel Studio got high marks during beta. >http://p.sf.net/sfu/intel-sw-dev >_______________________________________________ >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 ------------------------------------------------------------------------------ Download Intel® Parallel Studio Eval Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta. http://p.sf.net/sfu/intel-sw-dev _______________________________________________ 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
