On Tue, Jan 10, 2006 at 10:16:23AM -0800, Jesse Brandeburg wrote: >On Mon, 9 Jan 2006, Robin Humble wrote: >>until we turned off tso on our cluster using >> ethtool -K eth0 tso off >> ethtool -K eth1 tso off ... >>the major problems only happen for >32 cpu parallel runs. smaller runs >>work fine. unfortunately we haven't found a simple small MPI code that >>triggers the tso problems. >do you know what packet size triggered the problem? It sounds like the
unfortunately not really. >network traffic at the time of failure is lots and lots of outstanding >transmits over many concurrent connections, is that correct? repeatedly typing 'netstat -t' the most traffic I see is something like: [EMAIL PROTECTED] ~]# netstat -t | grep -v ' 0 0 ' Active Internet connections (w/o servers) Proto Recv-Q Send-Q Local Address Foreign Address State tcp 21720 0 beer96:57747 beer82:42185 ESTABLISHED tcp 37176 0 beer96:57747 beer76:45725 ESTABLISHED tcp 37176 0 beer96:38197 beer76:39912 ESTABLISHED tcp 37176 0 beer96:57747 beer86:54898 ESTABLISHED tcp 24 0 beer96:57747 beer74:57982 ESTABLISHED tcp 24 0 beer96:38197 beer74:59163 ESTABLISHED tcp 37176 0 beer96:57747 beer86:54883 ESTABLISHED tcp 24 0 beer96:57747 beer74:57967 ESTABLISHED tcp 24 0 beer96:38197 beer74:59178 ESTABLISHED tcp 37176 0 beer96:38197 beer82:34102 ESTABLISHED tcp 37176 0 beer96:38197 beer82:34117 ESTABLISHED tcp 0 37176 beer96:38197 beer86:48448 ESTABLISHED tcp 37176 0 beer96:57747 beer84:54840 ESTABLISHED tcp 37176 0 beer96:57747 beer84:54827 ESTABLISHED tcp 37176 0 beer96:38197 beer80:42088 ESTABLISHED tcp 17376 0 beer96:57747 beer90:55069 ESTABLISHED tcp 21720 0 beer96:57747 beer90:55058 ESTABLISHED tcp 37176 0 beer96:57747 beer81:57843 ESTABLISHED tcp 37176 0 beer96:38197 beer91:51101 ESTABLISHED tcp 24792 0 beer96:57747 beer75:44797 ESTABLISHED tcp 37176 0 beer96:38197 beer77:54658 ESTABLISHED tcp 37176 0 beer96:57747 beer83:46847 ESTABLISHED tcp 0 23168 beer96:57747 beer83:46826 ESTABLISHED tcp 21248 976 beer96:57747 beer89:48842 ESTABLISHED tcp 24 0 beer96:38197 beer75:51657 ESTABLISHED tcp 24 0 beer96:57747 beer73:52901 ESTABLISHED tcp 24 0 beer96:57747 beer73:52886 ESTABLISHED tcp 48 0 beer96:38197 beer73:47969 ESTABLISHED [EMAIL PROTECTED] ~]# there's about 140 sockets open total, but the rest have no traffic in them at this instant. >>we'd like to use tso as it means 5 to 10% less cpu usage for large >>message sizes (but strangely a few more micro-seconds latency). >>see attached pic. >Thats what TSO is supposed to help. The latency increase can be played >with or mitigated by changing tcp_tso_win_divisor in /proc/.../ipv4 cool. thanks. >>so what's the best way I can help you debug TCP segmentation offload >>issue? >we can start with getting some transmit ring dumps at the time of failure. >I have code to do this but need to port it to 2.6.15. i'll try to get >that code to you in the next couple of days. ok. ta. actually the tx reset only happens occasionally with the 2.6.15 kernel and 6.3.9 e1000 driver. mostly the code just stops - presumably 'cos a message got lost. I'll check more if it's at a repeatable place... cheers, robin - To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
