Hi,
we are currently testing load and failover in an ACI Multi-Pod fabric using Spirent test center. Round about 1000 devices are emulated sending traffic back and forth.
Everytime we start the test from scratch some of the flows (random ones, not always the same) show high amounts of packet drops. These don't show up in the interface counters of the fabric nor do they create any faults, but it's still a high amount (> 0.1 %). If the test is restarted after some time, or the flows are started sequentially rather than all at the same time, this phenomenon doesn't occur. Thus our assumption is, that this has something to do with MAC-Learning or ARP in the fabric. When too many new end points are presented to the fabric at one time, this seems to overload it.
Now I'm pretty sure we're not the first ones testing ACI with Spirent. Is there anyone recalling such problems and may be able to provide us some hints on how to get rid of the packet drops? Are there best practices for Spirent testing with ACI?
Bridge Domains are configured as default (now ARP Flooding, Hardware Proxy for Unknown Unicast), all Spirent ports are directly connected to the fabric, leaf switches are default gateway for the networks, there is no L2/L3Out whatsoever. ACI version is 3.1(1i).
Kind regards,
Nik
