Minutes for ROLL at IETF-94
minutes-94-roll-1

Agenda IETF 94

1hr Scheduled

10 minutes: WG Status

  Milestones, one

30 minutes: draft-robles-roll-useofrplinfo-02
10 minutes: draft-thubert-roll-dao-projection-02
10 minutes: Open Floor

[17:40]

Slides are https://www.ietf.org/proceedings/94/slides/slides-94-roll-3.pdf

Agenda approved.

Pascal Thubert draft-thubert-roll-dao-projection-02
extension to RPL for centralized routing computation.
uses DAO to establish route
at Detnet,
mechanism to populate the routes in the router unspecified at Detnet, could be
DAO for RPL routers. Last time, presented non-storing version. Message goes up
to the root and down. This time, storing mode. May optimize route a bit (to
common lowest ancestor). Still not very good. Proposal "projects" a DAO
straight into intended routers to install "shortcuts". Nodes through which
shortcut goes through must be neighboring. Root is then informed of shortcut
(via DAO-ACK probably) Last time, injected state into routers to reduce number
of nodes needed in source routing headers. Questions on the list: is this a new
mode of operation? Carsten: would like to exchnage this for storing mode. Could
help us get rid of storing mode. Michael: is this a new mode or an optino on
non-storing mode? Carsten: not very important. MR: if new mode, we need to have
a "flag day"... MR: this is not in the charter. Does the group want to
recherter for this, otherwise will close.

Pascal: 6550bis to make it more readable
Carsten: constrainedcast would fit well in a new charter.

[18:00] Michael draft-robles-roll-useofrplinfo-02
Reference diagram. Set of rules of encapsulation
Ramifications of rules not obvious.
Carsten:
MR: flow label discussed as a solution to IP-in-IP
All cases discussed during interim meeting Sept 29.
Describes a supposedly "simple" case of smart metering. The collection box
needs to know RPL otherwise drops the packet. Robert Cragie: Back to case: need
IP in IP and decapsulate at LBR. Leaf node needs to know destination is outside
LLN. Carsten: use "host" for things that are not routers ("non-RPL-aware Leaf"
in the slides), and "leaves" for routers that don't forward. Storing mode also
has problems. Two possible solutions: Robert: we did all this analysis with
Zigbee IP. Solution is to undo and redo IP in IP at every hop. And no IPIP from
last router to host. MR: 3 conclusions. Getting rid of storing mode solves a
lot of pbs; ??  ;?? Pascal: storing mode implemented in some systems. Pascal:
question at 6lo whether 6LoRH support... Carsten: would like to have a corpus
of packets that I can run to check performance of compressor variants. List of
packets and probability of occurence. MR: can we get this from Zigbee IP?
Robert Cragie: yes. In Zigbee, we concluded that in most cases, we have to
encapsulated. Minor inefficiency. MR: RPL to RPL storing mode is the one case
you don't have to encapsulate. Carsten: would recommend to always encapsulate,
and provide way to make it efficeint. RC: totally agreed. Destination is next
hop, gets full elision. Pascal: 6LoRH compresses outer IP very well. Pascal: in
most cases, the source doesn't know who will be the decapsulator. MR: drawback
is if we run over some non-lowpan network, we don't use 6LoRH, therefore don't
get this benefit.

[18:34]

Dominique - new draft on mailing list (draft-zhong-roll-dis-modifications-00).
Rewritten. Make the DIS messages more selected. Add flag into DIS and allow
metric container into DIS. Can control the responders so less traffic.
Typically for a node joining who knows nothing about the network and wants to
find routers of interest. Want routers to respond using unicast, not multicast.
Also add time dithering to avoid collisions from large number of routers
responding. If the flag is not set and options not present, everything works
like before. Does not break anything already existing.

Michael: This is looking for a node acting as assistant?
Dominique: Trying to find neighboring routers in an energy-efficient way