>On Tue, 2009-10-20 at 02:14 -0700, Carol Liu - Sun Microsystems - >Beijing China wrote: >> Hi the open-fcoe experts: >> I'm working on the FCoE initiator offload(performance) test on Linux >> with Intel ixgbe (Niantic 82599 driver). >> Could you guys give me some instruction about what kind of prerequisite >> must to be done for >> this offload(performance) test . >> >> Based on my knowledge , following is my steps to do the offload test. >> (eth4 is the ixgbe interface) >> 1. service dcbd start >> 2. ifconfig eth4 mtu 2500 >> 3. [r...@petpig ~]# dcbtool sc eth4 dcb on >> Version: 2 >> Command: Set Config >> Feature: DCB State >> Port: eth4 >> Status: Successful >> 4. fcoeadm -c eth4 > >I would run <open-fcoe>/debug/dcbcheck.sh to ensure that DCB is >configured correctly. > >Are you running the fcoe user space code? i.e. 'service fcoe start' >and modifying the /etc/fcoe/cfg-ethX files? > >Since you're using DCB you need fcoemon to be running to apply the >multiq qdisc and the required filters to direct FCoE traffic. > >These steps might help: > ># cp /etc/fcoe/cfg-ethx /etc/fcoe/cfg-eth4 <--- use whatever interface >you'll be doing fcoe on ># edit /etc/fcoe/cfg-eth4 to say- >FCOE_ENABLE="yes" >DCB_REQUIRED="yes" ># dcbcheck.sh eth4 (correct any reported failures) ># service fcoe start (assumes dcbd service is already started) > >> ----------------------------------------------------------- >> environment is : >> The target is Sun storageTek 6140 array >> FCoE switch is Cisco Nexus . > >This switch (depending on FW version) wants/requires you to use VLANs. > >> self-build kernel version : 2.6.31 i386 >> fcoe-utils: open-fcoe-2.6.30.tar.gz > >You might want to move to the 2.6.31 release. I'm not sure all of the >offload capabilities are supported in 2.6.30. > >> >> The performace data(by iometer/vdbench) I got after doing the steps >> above is not very well, so I was wondering if the dcbx ON the interface >> can 100% guarantee the hardware's offload is enabled ? or some other > >DCB being on is unrelated to offload support. In 2.6.31, FCoE offload is enabled when DCB is turned on, you must have DCB on to have FCoE offload on in 2.6.31.
> >> critical steps I missed ? what kind of way I can >> double check if the hardware's (Niantic ixgbe) offload is enabled or not ?. >> >> as far as I can see is one message from dmesg: >> ixgbe: eth4: ixgbe_set_fcoe_queues: FCOE enabled with DCB >> >> Really really appreciate any comments/instructions/suggestions you can >> offer me to help me move on , thanks ! >> >> /Carol Apart from what Rob suggested: First make sure your hw has offload cap turned on. Second, check you /sys/class/net/ethx/features: NETIF_F_FSO (1 << 21) must be set for fcoe sequence offload NETIF_F_CRC (1 << 24) must be set for FC CRC offload Third, before fcoeadm -c ethx, have fcoe debug on by (echo "0x0" back if there's too much in syslog): # echo "0x2" > /sys/module/fcoe/parameters/debug_logging You should be able to see messages like: Supports FCCRC offload. Supports LSO for max len 0x... Supports LRO for max xid 0x... Does "ethtool -S ethx | grep fcoe" give you fcoe stats? These would indicate if L2 type filter is setup right and working when you are passing FCoE traffic. If you have tshark/wireshark running on the initiator, you should not see FCP_DATA for SCSI read (read offload) but you should see as much as 32K frame size in FCP_DATA for SCSI write (sequence offload) depending on your write I/O size. yi _______________________________________________ devel mailing list [email protected] http://www.open-fcoe.org/mailman/listinfo/devel
