[ofa-general] QoS RFC

[Sean Hefty]

>> Application Service -> Service ID (or range)
>> Service ID -> desired QoS
>> QoS, SGID, DGID, PKey -> SGID, DGID, TClass, FlowLabel, PKey
>> SGID, DGID, TC, FL, PKey -> SLID, DLID, SL (set if crossing subnets)
>> SLID, DLID, SL -> MTU, Rate, VL, PacketLifeTime
> 
> What's the reasoning to use TClass and Flowlabel here? is it that you 
> want to come up with a scheme which is valid also for traffic crossing 
> IB subnets?

Yes - If you look at the data packet formats, the traffic class and flow 
label are the only fields that get carried between IB subnets.  They 
should be sufficient to specify a desired QoS.

In a similar fashion, the LRH carries the SLID, DLID, and SL, which 
should be sufficient to specify local subnet QoS.

> you suggest two mappings:
> 
> A) from SID to "desired QoS"
> B) from "desired QoS" (and more params) to TClass and FlowLabel
> 
> First, my understanding is that "QoS" is an abstract term here, that is 
> does not translate to concrete IB term, correct? now, what entity per 
> your design would translate from SID to QoS, is it something done 
> internally in the host stack similar to what the net stack does with the 
> IP_TOS and SOL_PRIORITY socket options?

Yes I view QoS as an abstract term, meaning we have a lot of flexibility 
in how it's defined.  I've chosen to define it wrt the hierarchy above. 
  Personally, I would drop the SID to QoS mapping, but if it is needed, 
keep the mapping done on the SA side.

- Sean