Wired Intelligent Edge

I manage a network of about 45 different HP2920 and 2530 switches across multiple campuses, and our QoS implementation has been nothing short of a nightmare due to the lack of detailed documentation on how this is actually implemented at the silicon level. If someone could let me know where I might be able to get good detailed information about how QoS works on these switches, that would be a huge help. If you're looking for info yourself, here's what I've been able to reverse engineer / establish on my own. I'm not sure how accurate this actually is, because like I said, I've found no documentation to support or deny this.

Both Switches:

Spanning-Tree and other critical management functionality all run through the QoS stack and take on whatever QoS queue the management VLAN has. If this queue overflows, your spanning-tree messages may be lost causing the network to diverge for a moment. This means for the best network stability you should dedicate one queue to network management.

Enabling QoS immediately partitions your buffer memory into equal parts per queue (so 4 queues gets 1/4 of the memory per queue, 8 queues 1/8th, etc.). This leads to buffer exhaustion under fairly routine circumstances as there's no place to "borrow" memory from.

HP2530

These switches always run QoS in WRR mode with tail-drop on the buffer.

HP2920

These switches can run in WRR or WRR with the highest priority being strict.

I think 4 queues puts a lot less strain on resources than 8. With 4 queues, the priority queue should be reserved by default for CoS 6 & 7. You then have to make sure you aren't allowing any applications to use 6 or 7. The "Voice VLAN" command unfortunately assigns 6 to voice, so you may need to remark, although voice doesn't use a lot of bandwidth so most people would never notice as WRR should clear the priority buffer just fine.

We have quite a lot of Aruba 2920 PoE switches with 16.03 (a few 16.04, I think).

Our IP phones are tagging voice traffic with a certain VLAN and marking voice packets with DSCP EF (46). So far so good.

On user switch ports we have configured trust DSCP and "voice" on the voice VLAN.

However, from looking at interface queue statistics, it seems clear that the switches are placing voice traffic in queue 8, which implies an 802.1p value of 7, since, according to "show qos queue-config," 802.1p 6 goes into queue 7.

Anyway we can deal with that easily.

What I really want to know is: how to ensure that queue 8 gets strict priority; and, more importantly, does "strict priority" mean that the switch services that queue whenever it has traffic (possibly starving other queues of service, even if only theoretically) or is it strict priority only up to its allowed bandwidth (this is how Cisco does priority queueing, to ensure that all queues get some service)?