Hi Donald,

> One thing I did notice was that the Receive side has 8 rx queues while the 
> Transmit has 16 rx queues.  You mentioned these were identical machines could 
> Hyper-Threading be off on the Receiving machine?
> 

Yes, it was my fault. I forgot to disable hyperthreading on TX server.
Now it is disabled, but the problem still remains.

> Also I'm still a little confused about how your test is set up.  If you have 
> four netperf's running on, as you called it, your Tx machine and your Rx is 
> only running netserver then I would only expect heavy traffic from Tx -> Rx.  
> Or am I missing some thing here?
> 

I try to summarize my tests.

For now I am testing only mono-directional TCP and UDP stream (from TX
server to RX server directly ). What I cannot understand is the low
throughput that I'm getting. The TCP tests shows a throughput of only
5.6 Gb/s. During UDP tests, with flow control disabled, I measured on TX
side a throughput of about 9.5 Gb/s (so the card can send at wire speed)
but on RX side I see only 5.6 Gb/s. I attribute this problem to the high
number of rx_missed_errors but I cannot figure out what is the real
cause (I don't know even what rx_missed_errors actually mean....).

Thanks

Mirko

>> -----Original Message-----
>> From: Mirko Corosu [mailto:[email protected]]
>> Sent: Tuesday, February 23, 2010 3:30 AM
>> To: Skidmore, Donald C
>> Cc: [email protected]
>> Subject: Re: [E1000-devel] Intel 2598EB 10-Gigabit AT dropped rx packet
>>
>> Hi Donald,
>>
>> Ok, sorry for the lack of informations: I'm not used to interact with
>> developers. :-)
>>
>>
>>> How were you using netperf in your tests?  What tests (I assume
>> [UDP|TCP]_STREAM) and how many instances of netperf were running each side?
>> I'm running four instances of UDP_STREAM or TCP_STREAM tests on the tx
>> side, with different packet sizes (from 1000 to 9000 byte) and a single
>> netserver on the rx side.
>>
>>> From the spec I got off the web on the R710 I can see they have either (2
>> PCIe x8 and 2 PCIe x4) or (1 PCIe x16 and 2 PCIe x4 ).  I was just
>> concerned you might be plugged into one of the x4 slots.
>>
>> I have  2 PCIe x8 on the so called "Riser 2" and 2 PCIe x4 on the "Riser
>> 1".   On each server the Intel cards are plugged into the Riser 2 slots.
>>
>>>>>  - Did you look at the dmesg for anything relative?
>>>> I see no error or warning messages on both side.
>>> It's not just error or warning messages I was interested in, although
>> they would be interesting too. :)  We also printk a message that displays
>> the type of PCIe link we have.  This would answer the question I posed
>> above about what slot the cards were in.
>> On rx side:
>>
>> [r...@grids2 ~]# dmesg |grep ixgbe
>> ixgbe: Intel(R) 10 Gigabit PCI Express Network Driver - version
>> 2.0.44.14-NAPI
>> ixgbe: 0000:07:00.0: ixgbe_init_interrupt_scheme: Multiqueue Enabled: Rx
>> Queue count = 8, Tx Queue count = 1
>> ixgbe: eth0: ixgbe_probe: (PCI Express:2.5Gb/s:Width x8) 00:1b:21:4b:c8:bf
>> ixgbe: eth0: ixgbe_probe: MAC: 1, PHY: 2, PBA No: d79893-017
>> ixgbe: eth0: ixgbe_probe: Internal LRO is enabled
>> ixgbe: eth0: ixgbe_probe: Intel(R) 10 Gigabit Network Connection
>> ixgbe: eth0: ixgbe_change_mtu: changing MTU from 1500 to 9000
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> None
>> ixgbe: eth0: ixgbe_remove: complete
>> ixgbe: Intel(R) 10 Gigabit PCI Express Network Driver - version
>> 2.0.44.14-NAPI
>> ixgbe: 0000:07:00.0: ixgbe_init_interrupt_scheme: Multiqueue Enabled: Rx
>> Queue count = 8, Tx Queue count = 1
>> ixgbe: eth0: ixgbe_probe: (PCI Express:2.5Gb/s:Width x8) 00:1b:21:4b:c8:bf
>> ixgbe: eth0: ixgbe_probe: MAC: 1, PHY: 2, PBA No: d79893-017
>> ixgbe: eth0: ixgbe_probe: Internal LRO is enabled
>> ixgbe: eth0: ixgbe_probe: Intel(R) 10 Gigabit Network Connection
>> ixgbe: eth0: ixgbe_change_mtu: changing MTU from 1500 to 9000
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> None
>> ixgbe: eth0: ixgbe_remove: complete
>> ixgbe: Intel(R) 10 Gigabit PCI Express Network Driver - version
>> 2.0.44.14-NAPI
>> ixgbe: 0000:07:00.0: ixgbe_init_interrupt_scheme: Multiqueue Enabled: Rx
>> Queue count = 8, Tx Queue count = 1
>> ixgbe: eth0: ixgbe_probe: (PCI Express:2.5Gb/s:Width x8) 00:1b:21:4b:c8:bf
>> ixgbe: eth0: ixgbe_probe: MAC: 1, PHY: 2, PBA No: d79893-017
>> ixgbe: eth0: ixgbe_probe: Internal LRO is enabled
>> ixgbe: eth0: ixgbe_probe: Intel(R) 10 Gigabit Network Connection
>> ixgbe: eth0: ixgbe_change_mtu: changing MTU from 1500 to 9000
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>> ixgbe: eth0: ixgbe_remove: complete
>> ixgbe: Intel(R) 10 Gigabit PCI Express Network Driver - version
>> 2.0.44.14-NAPI
>> ixgbe: 0000:07:00.0: ixgbe_init_interrupt_scheme: Multiqueue Enabled: Rx
>> Queue count = 8, Tx Queue count = 1
>> ixgbe: eth0: ixgbe_probe: (PCI Express:2.5Gb/s:Width x8) 00:1b:21:4b:c8:bf
>> ixgbe: eth0: ixgbe_probe: MAC: 1, PHY: 2, PBA No: d79893-017
>> ixgbe: eth0: ixgbe_probe: Internal LRO is enabled
>> ixgbe: eth0: ixgbe_probe: Intel(R) 10 Gigabit Network Connection
>> ixgbe: eth0: ixgbe_change_mtu: changing MTU from 1500 to 9000
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> None
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Down
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Down
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>> ixgbe: eth0: ixgbe_remove: complete
>> ixgbe: Intel(R) 10 Gigabit PCI Express Network Driver - version
>> 2.0.44.14-NAPI
>> ixgbe: 0000:07:00.0: ixgbe_init_interrupt_scheme: Multiqueue Enabled: Rx
>> Queue count = 8, Tx Queue count = 1
>> ixgbe: eth0: ixgbe_probe: (PCI Express:2.5Gb/s:Width x8) 00:1b:21:4b:c8:bf
>> ixgbe: eth0: ixgbe_probe: MAC: 1, PHY: 2, PBA No: d79893-017
>> ixgbe: eth0: ixgbe_probe: Internal LRO is enabled
>> ixgbe: eth0: ixgbe_probe: Intel(R) 10 Gigabit Network Connection
>> ixgbe: eth0: ixgbe_change_mtu: changing MTU from 1500 to 9000
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Down
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Down
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Down
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Down
>> ixgbe: eth0: ixgbe_remove: complete
>> ixgbe: Intel(R) 10 Gigabit PCI Express Network Driver - version
>> 2.0.44.14-NAPI
>> ixgbe: 0000:07:00.0: ixgbe_init_interrupt_scheme: Multiqueue Enabled: Rx
>> Queue count = 8, Tx Queue count = 1
>> ixgbe: eth0: ixgbe_probe: (PCI Express:2.5Gb/s:Width x8) 00:1b:21:4b:c8:bf
>> ixgbe: eth0: ixgbe_probe: MAC: 1, PHY: 2, PBA No: d79893-017
>> ixgbe: eth0: ixgbe_probe: Internal LRO is enabled
>> ixgbe: eth0: ixgbe_probe: Intel(R) 10 Gigabit Network Connection
>> ixgbe: eth0: ixgbe_change_mtu: changing MTU from 1500 to 9000
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>>
>>
>>
>> On tx side:
>>
>> [r...@client20 ~]# dmesg |grep ixgbe
>> ixgbe: Intel(R) 10 Gigabit PCI Express Network Driver - version
>> 2.0.44.14-NAPI
>> ixgbe: 0000:07:00.0: ixgbe_init_interrupt_scheme: Multiqueue Enabled: Rx
>> Queue count = 16, Tx Queue count = 1
>> ixgbe: eth0: ixgbe_probe: No DCA provider found.  Please start ioatdma
>> for DCA functionality.
>> ixgbe: eth0: ixgbe_probe: (PCI Express:2.5Gb/s:Width x8) 00:1b:21:4b:c8:e0
>> ixgbe: eth0: ixgbe_probe: MAC: 1, PHY: 2, PBA No: d79893-017
>> ixgbe: eth0: ixgbe_probe: Internal LRO is enabled
>> ixgbe: eth0: ixgbe_probe: Intel(R) 10 Gigabit Network Connection
>> ixgbe: eth0: ixgbe_change_mtu: changing MTU from 1500 to 9000
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> None
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Down
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Down
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Down
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>> ixgbe: eth0: ixgbe_remove: complete
>> ixgbe: Intel(R) 10 Gigabit PCI Express Network Driver - version
>> 2.0.44.14-NAPI
>> ixgbe: 0000:07:00.0: ixgbe_init_interrupt_scheme: Multiqueue Enabled: Rx
>> Queue count = 16, Tx Queue count = 1
>> ixgbe: eth0: ixgbe_probe: No DCA provider found.  Please start ioatdma
>> for DCA functionality.
>> ixgbe: eth0: ixgbe_probe: (PCI Express:2.5Gb/s:Width x8) 00:1b:21:4b:c8:e0
>> ixgbe: eth0: ixgbe_probe: MAC: 1, PHY: 2, PBA No: d79893-017
>> ixgbe: eth0: ixgbe_probe: Internal LRO is enabled
>> ixgbe: eth0: ixgbe_probe: Intel(R) 10 Gigabit Network Connection
>> ixgbe: eth0: ixgbe_change_mtu: changing MTU from 1500 to 9000
>> ixgbe: eth0: ixgbe_remove: complete
>> ixgbe: Intel(R) 10 Gigabit PCI Express Network Driver - version
>> 2.0.44.14-NAPI
>> ixgbe: 0000:07:00.0: ixgbe_init_interrupt_scheme: Multiqueue Enabled: Rx
>> Queue count = 16, Tx Queue count = 1
>> ixgbe: eth0: ixgbe_probe: No DCA provider found.  Please start ioatdma
>> for DCA functionality.
>> ixgbe: eth0: ixgbe_probe: (PCI Express:2.5Gb/s:Width x8) 00:1b:21:4b:c8:e0
>> ixgbe: eth0: ixgbe_probe: MAC: 1, PHY: 2, PBA No: d79893-017
>> ixgbe: eth0: ixgbe_probe: Internal LRO is enabled
>> ixgbe: eth0: ixgbe_probe: Intel(R) 10 Gigabit Network Connection
>> ixgbe: eth0: ixgbe_change_mtu: changing MTU from 1500 to 9000
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>> ixgbe: eth0: ixgbe_remove: complete
>> ixgbe: Intel(R) 10 Gigabit PCI Express Network Driver - version
>> 2.0.44.14-NAPI
>> ixgbe: 0000:07:00.0: ixgbe_init_interrupt_scheme: Multiqueue Enabled: Rx
>> Queue count = 16, Tx Queue count = 1
>> ixgbe: eth0: ixgbe_probe: No DCA provider found.  Please start ioatdma
>> for DCA functionality.
>> ixgbe: eth0: ixgbe_probe: (PCI Express:2.5Gb/s:Width x8) 00:1b:21:4b:c8:e0
>> ixgbe: eth0: ixgbe_probe: MAC: 1, PHY: 2, PBA No: d79893-017
>> ixgbe: eth0: ixgbe_probe: Internal LRO is enabled
>> ixgbe: eth0: ixgbe_probe: Intel(R) 10 Gigabit Network Connection
>> ixgbe: eth0: ixgbe_change_mtu: changing MTU from 1500 to 9000
>> ixgbe: eth0: ixgbe_watchdog_task: NIC Link is Up 10 Gbps, Flow Control:
>> RX/TX
>>
>>
>>
>>> All that said we would expect to see rx_missed_errors with flow control
>> disabled.  Since nothing would be stopping packets from overrunning the
>> buffer.  Your test seems to be agreeing with this as the number of
>> rx_missed_errors goes down with TCP (like you said) most likely due to TCP
>> sliding window.
>>> You don't see this issue with flow control enabled, right?
>> Right, but the throughput remains about 5.6 Gb/s and a can't figure out
>> where the bottleneck could be....
>>
>> Thanks a lot
>>
>> Mirko
>>
>>>> -----Original Message-----
>>>> From: Mirko Corosu [mailto:[email protected]]
>>>> Sent: Saturday, February 20, 2010 10:50 AM
>>>> To: Skidmore, Donald C
>>>> Cc: [email protected]
>>>> Subject: Re: [E1000-devel] Intel 2598EB 10-Gigabit AT dropped rx packet
>>>>
>>>> Hi Donald,
>>>>
>>>> Thank you for your reply.
>>>>
>>>>> I have a couple additional questions
>>>>>  - What were you running to do this test? (i.e. netperf, pktgen?)
>>>> netperf-2.4.2-1.el5.rf
>>>>
>>>>>  - What throughput were you seeing?
>>>> On tx side i see almost 10Gb/s, on rx side I see about 5.6 Gb/s
>>>>
>>>>>  - Are both ports plugged into x8 PCIe slots
>>>> Yes, both servers are Dell R710 with the same hardware configuration
>>>>
>>>>>  - Did you look at the dmesg for anything relative?
>>>> I see no error or warning messages on both side.
>>>>
>>>>>  - Do you see this issue only with UDP?
>>>> If I send a TCP stream I see on both side a throughput of about 5.54
>>>> Gb/s and less rx_missed_errors on rx side:
>>>>
>>>> NIC statistics:
>>>>      rx_packets: 1768101
>>>>      tx_packets: 235384
>>>>      rx_bytes: 15937257320
>>>>      tx_bytes: 15704254
>>>>      lsc_int: 2
>>>>      tx_busy: 0
>>>>      non_eop_descs: 7072222
>>>>      rx_errors: 0
>>>>      tx_errors: 0
>>>>      rx_dropped: 0
>>>>      tx_dropped: 0
>>>>      multicast: 1
>>>>      broadcast: 0
>>>>      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: 1539
>>>>      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: 1768098
>>>>      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: 1760417
>>>>      lro_flushed: 375277
>>>>      lro_recycled: 1012791
>>>>      rx_no_dma_resources: 0
>>>>      hw_rsc_count: 0
>>>>      rx_flm: 0
>>>>      fdir_match: 0
>>>>      fdir_miss: 0
>>>>      tx_queue_0_packets: 235384
>>>>      tx_queue_0_bytes: 15704254
>>>>      rx_queue_0_packets: 2
>>>>      rx_queue_0_bytes: 120
>>>>      rx_queue_1_packets: 8
>>>>      rx_queue_1_bytes: 786
>>>>      rx_queue_2_packets: 301752
>>>>      rx_queue_2_bytes: 2719959436
>>>>      rx_queue_3_packets: 562919
>>>>      rx_queue_3_bytes: 5074118826
>>>>      rx_queue_4_packets: 7
>>>>      rx_queue_4_bytes: 726
>>>>      rx_queue_5_packets: 15
>>>>      rx_queue_5_bytes: 1526
>>>>      rx_queue_6_packets: 313901
>>>>      rx_queue_6_bytes: 2829476778
>>>>      rx_queue_7_packets: 589497
>>>>      rx_queue_7_bytes: 5313699122
>>>>
>>>> I guess it's because of TCP flow control.
>>>>
>>>> I remember that the same behavior was triggered by enabling flow control
>>>> on both cards (ethtool -A eth0 rx on tx on autoconf on) but I have to
>>>> wait till Monday to perform another run to test that configuration.
>>>>
>>>> Thanks,
>>>>
>>>> Mirko
>>>>
>>>>> 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&#174; 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&#174; Ethernet, visit
>>>>>> http://communities.intel.com/community/wired
>>>> --
>>>>
>>>> ------------------------------------------------------------------------
>> --
>>>> Mirko Corosu
>>>> Istituto Nazionale Fisica Nucleare
>>>> Via Dodecaneso 33
>>>> 16146 Genova
>>>> www.ge.infn.it
>>>> Phone +39 010 3536361
>>>> ------------------------------------------------------------------------
>> --
>>>> ------------------------------------------------------------------------
>> --
>>>> Se tutto sembra venirti incontro, probabilmente sei nella corsia
>> sbagliata.
>>>> ------------------------------------------------------------------------
>> --
>>
>> --
>>
>> --------------------------------------------------------------------------
>> Mirko Corosu
>> Network and system administrator
>> Computing Center
>> Istituto Nazionale Fisica Nucleare
>> Via Dodecaneso 33
>> 16146 Genova, Italy
>> http://www.ge.infn.it
>> --------------------------------------------------------------------------
>>
>>
>>
>> --------------------------------------------------------------------------
>> Se tutto sembra venirti incontro, probabilmente sei nella corsia sbagliata.
>> --------------------------------------------------------------------------
> 


-- 
--------------------------------------------------------------------------
Mirko Corosu
Network and system administrator
Computing Center
Istituto Nazionale Fisica Nucleare
Via Dodecaneso 33
16146 Genova, Italy
http://www.ge.infn.it
--------------------------------------------------------------------------

--------------------------------------------------------------------------
Se tutto sembra venirti incontro, probabilmente sei nella corsia sbagliata.
--------------------------------------------------------------------------



------------------------------------------------------------------------------
Download Intel&#174; 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&#174; Ethernet, visit 
http://communities.intel.com/community/wired

Reply via email to