https://communities.intel.com/thread/54600?start=0&tstart=0
I am acquiring 800Mb/sec+ multicast data in a RHEV 6.5 VM using an HP 560SFP+ (Intel 82599ES chip set) adapter in passthru mode (i.e. the NIC is presented natively to the VM). I am seeing incrementing rx_missed_errors (and losing data) but rx_no_buffer_count is NOT incrementing. I see some Internet references that indicate that this might be because I have insufficient PCI bandwidth, but that does not seem likely because the platform is e5-2680 and I think I have dedicated PCI-e lines right to the CPU. I am using the stock RHEL 6.5 ixgbe driver 3.15.1k. My RX ring buffer is set to 4096, rx-usecs is 180, and lscpi indicates that I am connected at 66Mhz. Does anyone know what the incrementing rx_missed_errors is trying to tell me and is there any mitigation?
根据之前的经验总结， 一般情况下，如果看到了不少 rx_no_buffer_count，则表明需要增加 ring buffer；而如果看到 tx_deferred_ok 表示传输被延迟，不算 errors； rx_missed_errors, rx_fw_discards 这个表示包满了，多出的全被 drop 掉了。
http://jaseywang.me/2013/11/02/10g82599eb-%E7%BD%91%E5%8D%A1%E6%B5%8B%E8%AF%95%E4%BC%98%E5%8C%96ethtool/