So I'm just going to throw this out their and let anyone try this so that its known.
So the idea of Queue Egress Shaping with Bandwidth Egress Shaping Rates is not a bad idea but the way the SG300-10 does it can be a bit of a problem even with Strict Priority on 4 or more WRR Weight. The setup to test in short is your CIR Bandwidth Egress Shaping Rates = your Queue 4 Egress Shaping with Queue 3 Egress Shaping half the CIR of the bandwidth.
What I'm seeing and going by the Queues Statistics is packets are slowed in Queue 3 Egress Shaping but traffic going into Queue 4 Egress Shaping thats not a lot to exceed the total Bandwidth is having packet loss in Queue 4.
So here is what I think is happing which is there is only one buffer for all the Queues to Egress each Queue at different CIR but that buffer is getting full as packets come in for Queue 3 and can cause packet loss to happen in Queue 4 even if set to Strict.
But then I increased the Egress Committed Burst Size for the Bandwidth and now there is less packet loss happening in Queue 4. which means the WRR Weight is based on the Burst size for packets coming in but still in one buffer and packet loss can still happen in Queue 4 even if set to Strict.
I can guess at the reason for this which is lots more buffers with each buffer needing to be big enough to handle a high rate. But would it be possible to split the buffer into many buffers but with each split for a Strict Queue buffer the CIR & CBS for the bandwidth and Egress Shaping per Queue can only be set to half for each Strict Queue buffer.
1 Strict Queue buffer WRR with 3 Queues WRR in one buffer
CIR upto 500000 kbits/sec
CBS upto 8381451 Bytes
2 Strict Queue buffers WRR with 2 Queues WRR in one buffer
CIR upto 333333 kbits/sec
CBS upto 5587634 Bytes
The idea is the WRR is done across the buffers but with Strict Queue buffers not mixed in one buffer for packets to be in on sent into so that packet loss for Strict Queue buffers happens in its buffer and not under one buffer.
