Hi,
I'd just like to bring KIRA to your attention that was recently published
at IFIP Networking 2022: https://ieeexplore.ieee.org/document/9829816
or if you don't have access, you can also use the preprint version here:
https://publikationen.bibliothek.kit.edu/1000148953
[sorry for similar cross-posting from rtgwg, but this time with focus on
ANIMA use].
KIRA was designed to provide an extremely robust control plane
connectivity,
also for in-band control, so it is a "connectivity first" protocol that
tries to
uphold the connectivity between all its resources. In the ANIMA context,
it could
be seen as being an alternative to RPL in the ACP. In comparison to
RPL it does not create traffic concentrations and is very robust even
in drastic failure scenarios. Moreover, it can provide a built-in
DHT which could ease the discovery of ASAs and provide lightweight
lookup services.
Some features:
* It consists of a highly scalable ID-based routing protocol R²/Kad in
the routing tier
o highly scalable means 100,000s of nodes in a single domain
o "ID-based" means that it works on flat identifiers that have no
topological meaning, e.g., they could be hashes of public keys
or just random numbers
* it is a partially reactive path-vector protocol, i.e., a node
maintains a set of routes to some destinations, whereas it needs to
discover routes to other destinations on demand.
* It is completely self-organized (esp. zero-touch, zero-config)
* It is loop-free, even during convergence
* It shows good performance in various topologies (which we call
topological versatility), e.g., also in denser structures like data
center topologies.
* It achieves a good average stretch although its routing tables are
growing with O(log n) only (n=number of existing nodes in the network)
o Entries in the routing tables are shortest path routes
o Stretch is configurable by a node individual adaptation
mechanism, i.e., a node may achieve less stretch by providing
more memory for routing table entries. For example, an ASA may
put other ASAs it communicates frequently with into its routing
table.
* KIRA also provides a fast-forwarding scheme using PathIDs in the
forwarding tier
* R²/Kad routing protocol messages use source routing, whereas control
packets (i.e., ACP packets) forwarded by KIRA should use less
per-packet overhead and thus use a label-based forwarding scheme
that also supports multi-path forwarding.
o Currently, we use GRE encapsulation, but other methods could be
used, e.g., IPv6 SRH.
* The scheme currently uses IPv6 packets and ULA addresses, so it
would fit nicely into the ANIMA work.
We think that these features would make it a great choice as an ACP
routing protocol, however, it is clear that KIRA is nothing that has
been standardized yet.
Besides the simulation that was used to investigate KIRA's scalability,
we have a prototypical implementation as node-local SDN app that
provides IPv6 connectivity between the nodes (using OpenvSwitch), a
Linux-based native implementation is currently being developed. If
there is interest and agenda time available, I could try to give a brief
overview presentation at IETF 115, however, I understand that ANIMA
charter items have priority.
Regards,
Roland
_______________________________________________
Anima mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/anima
