Hi, I have an Intel X710DA2 NIC in a backup server I maintain at home. The server runs Gentoo Linux x86_64 and is kept pretty up to date - has been running gentoo 4.4.x kernels for some time now.
The server is used to back up around 1000G worth of data each week. This takes the form of a wake on LAN packet sent to the system which then wakes up, boots up and pulls down the data via scp before shutting down again. The data takes the form of an 880G tar file and some other smaller tarballs. The servers are in slightly different locations but are connected via a back-to-back 10G OM4 fibre with Intel optics end to end. The remote end has an X520 card and is running ESXi 6. Using the standard in-tree kernel i40e driver 1.3.46 I am able to get fairly good throughputs and able to transfer this data in about 90 minutes without stopping. However with the latest released i40e driver version 1.4.25 (from https://sourceforge.net/projects/e1000/files/i40e%20stable/) the backup job always fails. Somewhere after the transfer starts - perhaps 100G or so in - the connectivity falls over and the scp session doing the transfer fails . The job can be restarted again but it then fails again somewhere mid way through the transfer. The backup job logs this: /bin/tar: Removing leading `/' from member names Total bytes written: 90357760 (87MiB, 2.6MiB/s) packet_write_wait: Connection to 2001:44b8:xxxx:xxxx::2: Broken pipe lost connection /bin/tar: Removing leading `/' from member names packet_write_wait: Connection to 2001:44b8:xxxx:xxxx::2: Broken pipe This problem *doesn't* occur with the in-tree kernel driver which leads me to believe there is an issue with the 1.4 kernel that is posted online. Here's the dmesg output from the in-tree kernel driver: i40e: Intel(R) Ethernet Connection XL710 Network Driver - version 1.3.46-k i40e: Copyright (c) 2013 - 2014 Intel Corporation. i40e 0000:07:00.0: fw 5.0.40043 api 1.5 nvm 5.02 0x80002282 0.0.0 i40e 0000:07:00.0: The driver for the device detected a newer version of the NVM image than expected. Please install the most recent version of the network driver. i40e 0000:07:00.0: MAC address: 68:05:ca:30:53:d0 i40e 0000:07:00.0: PCI-Express: Speed 8.0GT/s Width x8 i40e 0000:07:00.0: Features: PF-id[0] VSIs: 66 QP: 8 RX: PS RSS FD_ATR FD_SB NTUPLE PTP VEPA i40e 0000:07:00.1: fw 5.0.40043 api 1.5 nvm 5.02 0x80002282 0.0.0 i40e 0000:07:00.1: The driver for the device detected a newer version of the NVM image than expected. Please install the most recent version of the network driver. i40e 0000:07:00.1: MAC address: 68:05:ca:30:53:d1 i40e 0000:07:00.1: PCI-Express: Speed 8.0GT/s Width x8 i40e 0000:07:00.1: Features: PF-id[1] VSIs: 66 QP: 8 RX: PS RSS FD_ATR FD_SB NTUPLE PTP VEPA i40e 0000:07:00.1 enp7s0f1: renamed from eth1 i40e 0000:07:00.0 enp7s0f0: renamed from eth0 i40e 0000:07:00.0 enp7s0f0: NIC Link is Up 10 Gbps Full Duplex, Flow Control: None i40e 0000:07:00.0 enp7s0f0: changing MTU from 1500 to 9000 The connectivity is straight IP, no other traffic on the card, no FC, no VLANs, nothing. Just IP over 10G Ethernet with scp and an MTU of 9000. Can someone suggest what we can do to narrow this down and ideally fix this in the 1.4 driver? I can run this with the in-tree kernel driver with is fine for now but I guess this still should really be fixed in the latest released driver anyway. Thanks, Reuben ------------------------------------------------------------------------------ Transform Data into Opportunity. Accelerate data analysis in your applications with Intel Data Analytics Acceleration Library. Click to learn more. http://pubads.g.doubleclick.net/gampad/clk?id=278785111&iu=/4140 _______________________________________________ E1000-devel mailing list E1000-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/e1000-devel To learn more about Intel® Ethernet, visit http://communities.intel.com/community/wired
