Description of problem: Sending TCP traffic over a e1000e NIC (big packets) transmits at 33 MB/sec instead of 110 MB/sec. I can see that the CPU core dealing with the interrupts for the network interface is maxed out at 100%. Everything works fine with kernel-2.6.31.5-127.fc12.x86_64, i.e., I can send at the full line rate without maxing out the CPU dealing with interrupts. The hardware is a 82571PT Intel NIC in a Sun X6270 blade server. Version-Release number of selected component (if applicable): kernel-2.6.33-0.18.rc4.git7.fc13.x86_64 How reproducible: Steps to Reproduce: 1. Send 1 Gbps traffic over TCP to a machine with a e1000e NIC. 2. 3. Actual results: Throughput is about 33 MB/sec. Expected results: Throughput should be 100 MB/sec. Additional info:
is TSO or GSO enabled? ethtool -k ethX
With the 2.6.31 kernel ethtool -k says: Offload parameters for e1000g0: Cannot get device flags: Operation not supported rx-checksumming: on tx-checksumming: on scatter-gather: on tcp-segmentation-offload: on udp-fragmentation-offload: off generic-segmentation-offload: on generic-receive-offload: off large-receive-offload: off With the 2.6.33rc kernel it says: Offload parameters for e1000g0: rx-checksumming: on tx-checksumming: on scatter-gather: on tcp-segmentation-offload: on udp-fragmentation-offload: off generic-segmentation-offload: on generic-receive-offload: off large-receive-offload: off So interestingly, on the old kernel, which works, there is that "Cannot get device flags" error.
Might be related to this bug reported on kernel.org: http://bugzilla.kernel.org/show_bug.cgi?id=15192
This bug appears to have been reported against 'rawhide' during the Fedora 13 development cycle. Changing version to '13'. More information and reason for this action is here: http://fedoraproject.org/wiki/BugZappers/HouseKeeping
I've tested the 2.6.33 and this bug has been resolved.
Based of above comment I'm closing the bug.