Hi, I’m working with some lab gear, current an array of 7048T-A and 7124SX switches, and trying to fully understand the hardware architecture to implement efficient enterprise-ish queuing mechanisms for converged infrastructures for pq/wrr-ish behavior.
Problem is, especially on the 1g, replacing cisco’s that are traditionally the trust edge boundary seems impossible when actually coloring traffic, as class map functionality seems unsupported, and using the a dumb “qos dscp x” is simply not suitable with something like a truly untrusted server for a dmz-ish environment. Main idea is to queue management/control-plane-ish traffic, and provide pq for media. Sysadmins look at me funny when I tell them to mark packets with dscp at a hypervisor level themselves, and this doesn’t seem to be changing converged ethernet or not.
Is there _any_ way to enforce untrusted port l3/l4 tuple-match coloring for setting dscp on port input at all or are markings assumed by arista to be done entirely on hosts for coloring?
Also, is there a good description how to utilize tail-drop behavior at the platform asic level between the various chips, the voq’s, vsq’s, and etc, how those relate to data vs. control-plane, traffic-queues, etc? I’ve not seen anything public or non that gives a good deep-dive on arista qos, about as close is Gary’s Arista Warrior, but still leaves much for someone prodding in catalyst and nexus qos for 15 years.
Thanks in advance!
Post your Answer
You must be logged in to post an answer.