>-----Original Message----- >From: Андрей Василишин [mailto:a.vasilis...@kpi.ua] >Sent: Wednesday, September 24, 2014 4:59 AM >To: Tantilov, Emil S; e1000-devel@lists.sourceforge.net >Subject: Re: [E1000-devel] Big paccket loss ixgbe, 3.22.3 > >Hello, Emil! > >24.09.2014 0:57, Tantilov, Emil S пишет: >> Were you not getting any rx_missed_errors before you >upgraded, or do you see more of them now? > >Before upgrade I got some rx_missed_errors, but after >upgrade I see them for 20-30 per second
It depends on the cause for the missed packets - it could be that the newer driver is simply faster for certain flows (see below). <snip> > >Now I using version: 3.6.7-k ># ethtool -S eth0 >NIC statistics: > rx_packets: 21467830008 > tx_packets: 39777419879 > rx_bytes: 5506957331678 > tx_bytes: 55752176077436 > rx_pkts_nic: 21534201436 > tx_pkts_nic: 39777419880 > rx_bytes_nic: 5597470666599 > tx_bytes_nic: 55911392882916 > lsc_int: 3 > tx_busy: 0 > non_eop_descs: 7314068 > rx_errors: 0 > tx_errors: 0 > rx_dropped: 0 > tx_dropped: 0 > multicast: 0 > broadcast: 58 > rx_no_buffer_count: 0 > collisions: 0 > rx_over_errors: 0 > rx_crc_errors: 0 > rx_frame_errors: 0 > hw_rsc_aggregated: 13893739 > hw_rsc_flushed: 2736888742 > fdir_match: 20661789898 > fdir_miss: 1830354720 > fdir_overflow: 5984 You are exceeding the number of flows Flow Director can support and seeing that you use FdirPballoc=3 for the driver there is not a whole lot you can do. The rx_missed_errors are most likely due to the constant clearing of the Fdir filters. The only other options are either try to make use of the perfect filters, or try to use receive packet steering (RPS from the kernel). <snip> >> 4. What type of traffic are you running? I understand that >this could be on a router or a system where you don't >control the traffic, but if you can find a common test (like >netperf) where you can show the increase in rx_missed_errors >that would be helpful. > >It is a cache server. Server traffic graph: >http://i.gyazo.com/461a14081d3c6cd04871489f0ecb0a61.png > >> 5. If you had flow control enabled - disable it and >collect the stats (ethtool -S) again. This can help identify >the issue better as you may see other counters increment >(such as rx_no_dma_resources). >> > ># ethtool -a eth0 >Pause parameters for eth0: >Autonegotiate: on >RX: on >TX: on > >root@host-81:~# ethtool -A eth0 autoneg off rx off You'll have to disable both tx and rx, but like I said your issue is most likely related to the number of flows exceeding the number of Fdir filters. I also noticed that you are getting lots of Rx checksum errors being reported, but again this could be related to the driver resetting the Fdir filters. Thanks, Emil ------------------------------------------------------------------------------ Meet PCI DSS 3.0 Compliance Requirements with EventLog Analyzer Achieve PCI DSS 3.0 Compliant Status with Out-of-the-box PCI DSS Reports Are you Audit-Ready for PCI DSS 3.0 Compliance? Download White paper Comply to PCI DSS 3.0 Requirement 10 and 11.5 with EventLog Analyzer http://pubads.g.doubleclick.net/gampad/clk?id=154622311&iu=/4140/ostg.clktrk _______________________________________________ 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
