Disclaimer
The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.
Liability Disclaimer
In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.
Posting
internal DSCP is an unclear concept and should not be considered
I respectively disagree. If the switch supports "internal marking" like DSCP's 6 bits, why not consider it? Of course not all switches offer DSCP like internal markings, 6500s internal COS (3 bits) comes to mind, but on other switches you can consider it.
There is a 4th state : conditionnal trust/marking with MQC but I will exclude the case
of course if you want to do QOS on non-IP traffic, DSCP is a useless concept - think pure L2 protocols like ARP or Q-in-Q);
You're correct concerning using IP ToS for the actual packet. But here's where the "internal" marking might come into use.
Consider if we you an explicit ingress policy, the "internal" QoS marking (hopefully DSCP like) might be used for QoS processing such a received packet.
There is no real "best effort" concept in the 3750; as soon as you enter the "mls qos" command in the config, a lot of underlying mechanisms are deployed (they don't appear as the switch use default value); for example; buffers on each interface are allocated at 25% for each queue at egress; the queue 1 is shape (rate limit+ BW guaranteed) at 25%; queue 2, 3 and 4use smooth round robin at 1/3 of the remaining bandwidth when queue 1 is full; otherwise if queue 1 is empty, queue 2, 3, 4 will share 1/3 (not limited) of the available bandwidth (so; all the link if q1 is empty)
Unclear what you mean with "There is no real 'best effort' concept in the 3750". What do you understand the "best effort" concept to be?
By default, you're correct when the 3750 QoS is enabled, 4 egress queues are enabled with default settings. However, part of the default QoS configuration is ingress markings are not trusted and are even erased. Effectively, all ingress traffic is treated alike, i.e. "best effort", although with the default buffer settings, the single "default marking" egress queue for all non-trusted traffic doesn't have the buffer resources the only egress queue has with QoS disabled.
BTW the 3750's default SRR share configuration of 25 0 0 0 is 1/25 (4%) of the bandwidth, it's not 25%. Although also by default a 3750's 4 egress queues are assigned 25% of the buffer space, the remaining default parameters usually don't reserve all this space for each egress queue and the unreserved buffer space may be acquired by other egress queues. I.e. (much) more than 25% might be actually actively (dynamically) allocated to an egress buffer. Lastly, with default shaping for queue 1, the queue doesn't have to be full to impact bandwidth sharing between Q1 and the other queues. With the usual default configuration, all 4 egress queues will share bandwidth equally except when Q1 traffic is being held due to shaping, then the other remaining queues will continue to share the bandwidth equally.
