On what basis would the mapping system decide if a full match or prefix
match is appropriate? So far, each EID has been specific on what kind
of lookup it does. IP (v4 or v6) lookups always do LPM lookups. All
other EIDs we have specified so far do exact matches.
Yours,
Joel
On 10/2/2020 11:22 AM, Dino Farinacci wrote:
The document looks to me underspecified.
I'm sorry, it is completely specified. I'll respond to a subjective comment
with a subjective reply.
There is no formal definition of “Distinguished Name”, for what is worth the
document can be renamed "LISP LCAF String encoding”
I provided a reference in the document to th Address-Family-Identifiers
document. Where it is listed. That creates the code point. And this document
says how LISP will use AFI=17.
The slides you presented during IETF 96 suggest that there some longest
characters match (like longest prefix match but on strings) and this part is
not documented in the draft.
You can do partial match lookups on strings. If a distinguished-name "luigi" is
registered as an EID to the mapping system, and a lookup is done on "luigi-iannone", the
map-server can decide if partial or exact matches are performed. I can make this more clear in the
document.
As an implementer, the fact that there is no explicit length field is
worrisome, performance wise.
Because if I need to skip through a LCAF AFI=17 record I have anyway to parse
the string since I do not know where it ends.
This can slow down operations.
Zero termination of the string creates the length for implementation that can
support AFI=17. For implementation that do not, it uses the EID mask-length to
skip over an EID in an EID-record. For RLOC-records if you don't understand an
AFI, you drop the packet.
I think the draft should include some specific use cases that prove that LCAF
strings are really useful.
Strings are useful. How they are encoded is another topic.
The ECDSA use case is not a compelling example since you can use LCAF AFI=XX
that identifies public keys encoded in a specific way (the WG should think
about proposing such encoding….)
That makes name uses more bytes in the packet. We want to encode a string of
length 5 in 8 bytes. 2 bytes for AFI and 1 byte null-character. You can encode
it more efficicently than that. And that is why AFI=17 was chosen.
Dino
_______________________________________________
lisp mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/lisp
