Re: [Isis-wg] Encoding inconsistency between ISIS and OSPFv2 extensions for SR

[Xuxiaohu <[e-mail]>]

Hi Peter,

Please see my response inline

> -----Original Message-----
> From: Peter Psenak [mailto:[e-mail]]
> Sent: Friday, June 13, 2014 8:31 PM
> To: Xuxiaohu; [e-mail] list; OSPF List
> Subject: Re: [Isis-wg] Encoding inconsistency between ISIS and OSPFv2
> extensions for SR
> 
> Hi Xiaohu,
> 
> please see inline:
> 
> On 6/13/14 12:09 , Xuxiaohu wrote:
> > Hi peter,
> >
> >> -----Original Message-----
> >> From: Isis-wg [mailto:[e-mail]] On Behalf Of Peter
> >> Psenak
> >> Sent: Friday, June 13, 2014 4:32 PM
> >> To: Xuxiaohu; [e-mail] list; OSPF List
> >> Subject: Re: [Isis-wg] Encoding inconsistency between ISIS and OSPFv2
> >> extensions for SR
> >>
> >> Xiaohu,
> >>
> >> please see inline:
> >>
> >> On 6/13/14 09:51 , Xuxiaohu wrote:
> >>> Hi all,
> >>>
> >>> There are some encoding inconsistencies between OSPFv2 extensions
> >>> and ISIS
> >> extensions for SR as follows:
> >>>
> >>> 1. In ISIS-SR, the prefix-sid advertisement is piggybacked on the IP
> >>> reachability
> >> advertisement. In OSPF-SR, the prefix-sid advertisement is
> >> piggybacked on OSPF Extended Prefix LSA which is used to advertise
> >> other attributes associated with the prefix, rather than the
> >> reachability. IMHO, the OSPF encoding is more flexible since the
> >> label distribution and the reachability advertisement are
> >> independent. As a result, the route summary on area boundaries at
> >> least can be enabled as before. Besides, the prefix-SID sub-TLV can
> >> be used to advertise a range of prefix/SID pairs (see item2). In
> >> fact, ISIS allows us to do the same way as OSPF with a much lower
> >> cost (see section 3 of
> http://tools.ietf.org/html/draft-xu-isis-global-label-sid-adv-00). Of course, it
> seems that you co-authors prefer to the piggyback way.
> >>
> >> OSPF LSAs that are used to advertise the prefixex are not extensible,
> >> so we had to define a new LSA for the purpose of advertising a prefix related
> attributes.
> >> ISIS is different, as they can add sub-TLVs to existing TLVs.
> >
> > I see. For ISIS, you use the piggyback way (piggyback the label/sid
> advertisement on the reachability advertisement messages). For OSPFv2, you
> have no way to use the piggyback way anymore, so you defined a new LSA...
> That's why I said you prefer to the piggyback way. However, I don't think the
> piggyback way is much worthwhile from the perspective of flexibility and
> extensibility.
> >
> >>>
> >>> 2. In ISIS-SR, the Prefix-SID Sub-TLV can only be used to advertise
> >>> an SID for a
> >> single prefix. And it relays on the SID/Label Binding TLV to
> >> advertise a range of prefix/SID pairs. In contrast, In OSPF-SR, the
> >> prefix-sid sub-TLV can be used to specify a range of addresses and
> >> their associated Prefix SIDs. By the way, in the 4.3.  SID/Label
> >> Binding sub-TLV. it has the following text: "Range Size: usage is the
> >> same as described in Section 4.2." Did you co-authors want to propose
> >> two ways (i.e., prefix-sid sub-TLV and SID-Label Binding sub-TLV) to achieve
> the same goal (i..e, advertise a range of prefix/SID pairs)?
> >>
> >> because in OSPF advertisement of the prefix SID is decoupled from the
> >> advertisement of prefix reachability, we can afford to advertise the
> >> range of SIDs in the prefix-SID sub-TLV as such.
> >
> > IMHO, the ISIS and OSPFv3 advertisement of the prefix SIDs should be
> > decoupled from the prefix reachability advertisement as well:)
> 
> in OSPFv3 case, we have a way to advertise the prefix using the proposed
> encoding in draft-acee-ospfv3-lsa-extend, but do not advertise the reachability
> of the prefix - it's call NU-bit (rfc5340, A.4.1.1.)

That's great. BTW, don't you believe the ISIS protocol has provided almost the same capability as the NU-bit (see the following text quoted from RFC5305)?

"...If a prefix is advertised
   with a metric larger then MAX_PATH_METRIC (0xFE000000, see paragraph
   3.0), this prefix MUST NOT be considered during the normal SPF
   computation.  This allows advertisement of a prefix for purposes
   other than building the normal IP routing table...".  

> >
> >> No, we do not define two ways to achieve the same thing. Binding TLV
> >> is used for a different purpose and the same usage is only applicable
> >> to the Range semantics, not to the whole Binding TLV.
> >
> > Does that mean the Binding sub-TLV in the OSPF-SR could not be used to
> advertise a range of prefix/sid pairs while the binding sub-TLV in the ISIS-SR
> could?
> 
> Binding TLV in OSPF is only used to advertise a "LSP path" local to the advertising
> router, it's not used for anything else. YOu can still advertise a single "LSP path"
> for range of prefixes.

Don't you believe it's better for the Binding TLV in ISIS to be used to advertise a LSP as well? 

> In ISIS, due to the need to decouple prefix reachability from SID advertisement,
> Binding TLV is used for SR Mapping Server (SRMS) adevrtisement on top of what
> it is used in OSPF (in OSPF SRMS advertisements are using the Prefix/SID
> sub-TLV).

To decouple prefix reachability from SID advertisement, why not consider the approach of using the MAX_PATH_METRIC trick (see section 3 of http://tools.ietf.org/html/draft-xu-isis-global-label-sid-adv-00)?

Best regards,
Xiaohu

> >>> 6. In ISIS-SR, the prefix-SID sub-TLV doesn't contain the MT-ID
> >>> field since the
> >> MT-ID field is already contained in the parent TLV of the prefix-SID
> >> sub-TLV. In OSPF, the MT-ID field is contained in the Prefix SID
> >> Sub-TLV since the parent TLV of the prefix-sid sub-TLV doesn't
> >> contain that MT-ID field. IMHO, it's better to contain the MT-ID in
> >> the parent prefix-specific TLV of the prefix-SID sub-TLV. In other
> >> words, why not contain the MT-ID in the OSPF Extended Prefix TLV,
> >> instead of the prefix-sid sub-TLV (see section 3 of
> http://tools.ietf.org/html/draft-xu-ospf-global-label-sid-adv-00)?
> >>
> >> no, we do not want to put the MT-ID in the OSPF Extended Prefix TLV.
> >> The reason is that attributes are MT specific, not the prefix itself - e.g.
> >> you may want to advertise different metrics for the same prefix in
> >> different topologies, not the same prefix twice.
> >
> > Make the prefix-sid as a sub-TLV of the Multi-Topology sub-TLV?
> 
> no, we don't want to end up with sub-sub-TLVs right from the beginning.
> 
> regards,
> Peter
> 
> >
> > Best regards,
> > Xiaohu
> >
> >> regards,
> >> Peter
> >>
> >>
> >>
> >>
> >>
> >>>
> >>> Anyway, although it is unavoidable for us to define extensions to
> >>> both ISIS and
> >> OSPF for the same thing due to the fact that both protocols have been
> >> widely used, could we try our best to keep the encodings of ISIS and
> >> OSPF as consistent as possible for the same functionality? In this
> >> way, once someone has read the ISIS extension draft, he or she can
> >> easily think of what has been done in the OSPF extension draft accordingly,
> and vice verse.
> >>>
> >>> Best regards,
> >>> Xiaohu
> >>>
> >>> _______________________________________________
> >>> Isis-wg mailing list
> >>> [e-mail]
> >>> https://www.ietf.org/mailman/listinfo/isis-wg
> >>> .
> >>>
> >>
> >> _______________________________________________
> >> Isis-wg mailing list
> >> [e-mail]
> >> https://www.ietf.org/mailman/listinfo/isis-wg
> > .
> >