Re: PQC and QOSPF

[Zhaohui Zhang <[e-mail]>]

Yao-Min,

Yes that's my understanding.

Jeffrey

> 
> Jeffrey, 
> 
> Thank you for the prompt and authoritative reply from 
> the QOSPF point of view. 
> 
> Summarizing from your response, RRA flooding is needed
> to avoid "stepping over your own shadow" but such 
> flooding indeed leads to a scaling problem.  Possible 
> approaches to alleviate the problem:
> - Let RRA info be propagated only to routers on the data 
>   path, possibly using RSVP.  But RSVP camp did not like 
>   the idea.  
> - Use ER, so that the info is only needed at the first 
>   hop routers.  
> 
> Using RSVP to carry routing related info does appear
> to violate the separation between reservation and 
> routing.  Now, if we use ER, still there needs to be a 
> way to propagate the resource reservation info to the 
> first-hop (source) routers.  For this purpose, 
> EROSPF (Explicit Routing OSPF) proposed in in 
> draft-zhang-qospf uses RRAs but limits the 
> flooding destination only to the first-hop 
> routers (but these RRAs have to be propagated 
> through intermediate routers anyway).  The 
> complexity analysis in the draft seems to indicate 
> the same message complexity as PQC.  Let me know 
> if this comparison is incorrect. Thanks. 
> 
> Yao-Min 
> 
> 
> 
> 
> Zhaohui Zhang wrote:
> > 
> > Yao-Min,
> > 
> > I have not read the paper yet, so my response is based your email only.
> > 
> > >
> > > A criticism on QOSPF made in the PQC paper
> > > (http://www.isoc.org/inet97/proceedings/F4/F4_2.HTM)
> > > was the scaling issue with the Resource Reservation
> > > Advertisements (RRAs).  RRA, being per-flow
> > > information, is flooded throughout an OSPF area.
> > >
> > > The approach proposed in PQC is to have the resource
> > > reservation information for a flow only propagated
> > > to routers that are on the path of the flow.
> > > The propagation mechanism uses RSVP.
> > 
> > Flooding RRAs anywhere does cause the scaling problem, but from day one
> > we proposed an alleviation - use Explicit Routing.
> > 
> > With ER, reservation information is sent only to the "first hop"
> > routers - even better than sending to routers on the path of the flow.
> > 
> > We also tried to put it into RSVP so that only routers on the path would
> > get the info, but RSVP folks did not think that was a good idea, so we
> > had to give up.
> > 
> > >
> > > Is there a compelling reason in the design of
> > > QOSPF to have RRAs flooded throughout an area?
> > 
> > The resource information is needed to deal with "step over your own
> > shadow" problem. Any router involved in the calculation of the path for
> > a flow needs to know the reservation info for the flow - I believe you
> > share the same view.
> > 
> > Flooding them anywhere does not scale and is not necessary, yet it is easy
> > to understand and simple to implement, so we did not put ER up front but
> > put it in the appendix instead.
> > 
> > >
> > > --
> > > Yao-Min Chen, Fujitsu Labs of America, [e-mail]
> > > 595 Lawrence Expressway, Sunnyvale, CA 94086-3922
> > > Phone: [phone] Fax: [phone]
> > >
> > 
> > Thanks.
> > 
> > Jeffrey
>