I'd also be inclined toward quirky 9k internals. I believe a colleague of mine troubleshot an issue with latency/slowness through some Nexus switches (I can't recall if they were 9ks). After engaging TAC, they noticed that "no ip redirects" was applied to the VLAN 1 SVI but none of the other SVIs. While it theoretically shouldn't have made any difference, they applied "no ip redirects" to the rest of the SVIs and everything started working just fine. Sincerely, Anthony Critelli B.S. Applied Networking and Systems Administration, 2014 www.acritelli.com (845) 283-4117 On Sun, Jul 24, 2016 at 12:32 PM, Jeremy <jbaino@gmail.com> wrote:

Running into some weird issues with a Cisco Nexus9k.

We have a Cisco 3750X pair stacked, port channel (2x 1G) to a two different blades on a Nexus9k. Isolating the links of the port channel , on one link we can consistently get 800mbps (using iperf), or the other link we consistently get ~34mbps.

we have seen this across multiple 3750X stacks.

The vlan we were on is just layer2 through the n9k, there are no IP addresses. We were able to (apparently) resolve this issue by creating an SVI on the n9k, with an empty config.

Now, even isolating links we can get ~800mbps across the n9k, through the various 3750X stacks.

I am confused why creating the SVI would have an impact on this, and why it wouldn't be consistent across both links. If the lack of SVI were at fault, I would be less surprised if it just flat out didn't work, but this partial working state feels very odd.

Anyone else seen this? Thoughts? Could traffic be hitting the CPU while going across modules? This feels like quirky n9k internals...

Thanks! Jeremy

PS: no CRC errors found on interfaces, all looked clean