Hi, Has anyone using CentOS 6 been able to successfully set an mtu larger than 9710 on an interface. I am seeing super jumbo frames with length > 10000. ... IP 10.79.4.53.64327 > 10.79.2.53.24294: Flags [.], seq 16060:29200, ack 1, win 32767, length 13140 ... CentOS release 6.7 (Final) Thanks, -- Stephen Clark *NetWolves Managed Services, LLC.* Director of Technology Phone: 813-579-3200 Fax: 813-882-0209 Email: steve.clark at netwolves.com http://www.netwolves.com
On 11/09/2015 08:34 AM, Steve Clark wrote:> > Has anyone using CentOS 6 been able to successfully set an mtu larger > than 9710 > on an interface.Maximum frame size varies from implementation to implementation: http://pages.uoregon.edu/joe/jumbo-clean-gear.html It's also worth noting that that due to offloading features in modern NICs, there's often very little benefit to large frames. Since you *really* need all of the devices on any network segment to use the same MTU, the best option might be to eliminate jumbo frames (and test the impact of doing so in terms of throughput and CPU utilization).
On 11/09/2015 12:36 PM, Gordon Messmer wrote:> On 11/09/2015 08:34 AM, Steve Clark wrote: >> Has anyone using CentOS 6 been able to successfully set an mtu larger >> than 9710 >> on an interface. > Maximum frame size varies from implementation to implementation: > http://pages.uoregon.edu/joe/jumbo-clean-gear.html > > It's also worth noting that that due to offloading features in modern > NICs, there's often very little benefit to large frames. Since you > *really* need all of the devices on any network segment to use the same > MTU, the best option might be to eliminate jumbo frames (and test the > impact of doing so in terms of throughput and CPU utilization). >Hi Gordon, Thanks for the response. The real issue relates to doing pcap on an interface that is hooked to a span port. The super jumbo frames cause rx_long_length_errors: 2701813 which show up in our monitoring software and the customer thinks there are error on his network. So I wanted to increase the mtu on the interface so these errors would not be reported. -- Stephen Clark *NetWolves Managed Services, LLC.* Director of Technology Phone: 813-579-3200 Fax: 813-882-0209 Email: steve.clark at netwolves.com http://www.netwolves.com
> On Nov 9, 2015, at 11:34 AM, Steve Clark <sclark at netwolves.com> wrote: > > IP 10.79.4.53.64327 > 10.79.2.53.24294: Flags [.], seq 16060:29200, ack 1, win 32767, length 13140Do you have RSS enabled? With RSS the software/tcpdump sees larger "packets" but the physical NIC chunks down to the wire MTU. What does a capture on the destination show? https://en.m.wikipedia.org/wiki/Receive-side_scaling#RSS