[Diffserv] RE: Queries on Diffserv MIB (RFC3289)

["Andrew Smith" <[e-mail]>]

Vikas,

Here are some strawman answers to your questions: I hope others will
correct me if I'm wrong.

1. I believe that the diffServTBParamType object determines whether the
conformance is "strict", "loose" or "whatever". The "standard" values
defined in the MIB (diffServTBParamSimpleTokenBucket,
diffServTBParamAvgRate, diffServTBParamSrTCMBlind,
diffServTBParamSrTCMAware, diffServTBParamTrTCMBlind,
diffServTBParamTrTCMAware and diffServTBParamTswTCM) all have a
REFERENCE clause that will lead you to some sort of definition of the
intended algorithm.

2. As I understand RIO, you would connect a Meter to two separate
Algorithmic Droppers, one for "in" and one for "out". You may also want
to run RIO on multiple classes (e.g. AF1x, AF2x) in which case you would
have a Classifier looking at the DSCPs, with its 2 (or more) outputs
each connecting into a {Meter -> 2 Algorithmic Droppers} block. [Or did
we just decide that RIO was out of scope? It's so long ago ...]

3. You are correct that the guidance in the RFC: "Changes in
[ThreshBytes] may or may not be reflected in the reported value of
[ThreshPkts]" is somewhat weak. I believe that the WG felt that an
implementation would probably "just know" whether it counted bytes or
packets and was unlikely to be able to support both. Maybe some
implementation feedback is needed on this one? 

4. I believe that the WG felt that Qthreshold was likely to represent a
finite chunk of memory, an upper bound on a queue length or buffer size,
usually measured in bytes. I agree that this might not always be true.
Again, implementation feedback will be useful here. I do not think it
necessarily needs to be directly correlated with whether a threshold is
in bytes or packets.

Hope that helps,

Andrew Smith

P.S. The usual advice of "go read the WG archive" would be unfair in
this case as there is so much diverse stuff in that archive.

-----Original Message-----
From: Vikas Bansal [mailto:[e-mail]] 
Sent: Friday, July 19, 2002 6:39 AM
To: [e-mail]
Subject: Queries on Diffserv MIB (RFC3289)


Dear Sir,

I need a couple of clarification regarding Diffserv MIB RFC3289. I'll
appreciate if you could respond at earliest.

1. The Informal Management Model (RFC3290) talks about two types of
conformance for the meter - strict and loose. Is there a way to specify
this in MIB or is it implementation dependent.

2. Multiple algorithmic droppers can be used to implement WRED by having
their diffServAlgDropQMeasure pointing to the same queue. But, if we
need to implement RIO algorithm then Algorithmic Dropper should take an
additional specific parameter for the drop priority level.

3. DiffServRandomDropEntry contains thresholds in both in terms of
number of bytes and number of packets. How should the entity decide
whether it should use byte threshold or packets threshold for the
dropping algorithm.

4. diffServAlgDropQThreshold is defined as "A threshold on the depth in
bytes of the queue ...". The unit of measurement defined is bytes. If
the dropper algorithm measures the queue depth in terms of number of
packets than the unit of diffServAlgDropQThreshold should also be the
number of packets. Can the same field be used in that case to specify
the threshold in number of packets.

Thanks and Regards
Vikas Bansal
Principal Software Engineer
GlobespanVirata, Inc.





_______________________________________________
diffserv mailing list
[e-mail]
https://www1.ietf.org/mailman/listinfo/diffserv
Archive: http://www.ietf.org/mail-archive/working-groups/diffserv/current/maillist.html