DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark...

46
DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland
  • date post

    22-Dec-2015
  • Category

    Documents

  • view

    215
  • download

    2

Transcript of DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark...

Page 1: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

DNA design team update

JinHyeock Choi

Tero Kauppinen

James Kempf

Sathya Narayanan

Erik Nordmark

Brett Pentland

Design Team:

Brett Pentland

Page 2: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Progress

• Identified and catalogued a range of solutions.• Discussions have centred on the pros and cons

of available techniques.• Next steps:

– Solicit more new ideas. – Develop of list of filtering criteria.– Apply filtering and choose a particular set of ideas. – Iron out the details.

Page 3: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Assumptions

• Layer-2 triggers to initiate RS message will be available.– RS delay is an issue?

• Routers on a link can hear each other.– Link = broadcast domain

• Hosts that can connect to multiple links from one interface can distinguish packets from the different links.

• DNA is complete when the link identity decision is made and the information required for IP configuration obtained.

Page 4: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

The Problem

• Checking for link change.– Putting the right information in the RS/RA

exchange so that an accurate decision can be made after reception of the first RA.

• Getting the RA fast.

Page 5: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Checking for Link Change

• Adding something to the RA:– Explicit Link Identifier

• Random• PIO based• Hashed Prefix based

– Complete RA

• Adding something to the RS to ask a question:– Prefix-based Landmark– Priority Landmark– Hybrid Landmark

Page 6: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Getting the RA quickly

• Fast Router Discovery (RA caching)

• Simple FastRA

• Deterministic Fast RA

• Hash-based deterministic Fast RA

• Probabilistic Fast RA

Page 7: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Checking for Link Change

Page 8: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Explicit Link Identifier - Random

• Agreement on a random number as an explicit link identifier between routers.– LinkID MUST be different from LinkID of all links from

which a host could directly transition from.

• Include this identifier in a new option in RA messages.

Page 9: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Explicit Link Identifier - Random

Pros• Easy for host to

recognise link reliably.• Doesn’t require

solicitation to be useful.• Independent of prefix

changes on link.• Single (sol/unsol) RA

will in one swoop help all hosts.

Cons• Establishing

agreement between routers.

• Ensuring uniqueness from adjacent links.

• Extra option in RAs.

Page 10: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

PIO (Prefix Information Option) Link Identifier

• Agreement on using a particular prefix as an explicit link identifier between routers.

• Try to find a prefix common to all routers.– If not possible, need some kind of option to add where

the LinkID is not a configured prefix.

Page 11: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Pros• Easy for host to

recognise link reliably.• Doesn’t require

solicitation to be useful.

• No extra options if there is a common prefix.

• Single (sol/unsol) RA will in one swoop help all hosts.

Cons• Needs agreement

between routers if single prefix is chosen.

• Link Identifier may need to change as prefixes on the link change.

PIO (Prefix Information Option) Link Identifier

Page 12: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

• Hash of all active prefixes on the link to create a Link Identifier.

• Include this identifier in a new option in RA messages.

Hashed Prefix Link Identifier

Page 13: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Hashed Prefix Link Identifier

Pros• Easy for host to

recognise link reliably.• Doesn’t require

solicitation to be useful.

• Single (sol/unsol) RA will in one swoop help all hosts.

Cons• All prefixes must be

known to all routers.– Any variation makes the

identifier unrecognisable

• Link Identifier needs to change as prefixes on the link change.

Page 14: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Complete RA

• Routers learn about all prefixes on the link.• Include all the prefixes on the link in RA

messages.– Flag to indicate completeness.– Prefixes not explicitly configured on router

recognizable as such.

Page 15: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Complete RA

Pros• Easy for host to

recognise link reliably. • Small RAs if all routers

on a link have the same prefixes.

• No changes to solicitations.

• Single (sol/unsol) RA will in one swoop help all hosts.

Cons• RAs carry extra

prefixes if routers have non-matching prefix sets.

• Potential for large increase in RA size.

• Synchronization if prefixes change (less stringent).

Page 16: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Requested landmark

• Routers learn about all prefixes on the link.• Hosts include a current prefix in their RS

message.• Responding routers include ‘yes/no’ flags to

indicate if that prefix is in use in the link.– The order of the responding routers is left to fastRA

schemes.

Page 17: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Requested Landmark

Pros• Simple RAs (yes/no

flags).• No router

negotiations.– Just listen to RAs.

Cons• Needs solicitation to be

useful.• Best suited for unicast RA.

Aggregation gets tricky.• Need to ensure all prefixes

are known to all routers – synchronization –(or) Choice of responding router could lead to erroneous conclusions.

• Prefix option added to RSs.

Page 18: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Priority landmark

• Hosts include their current prefix and current default router address in their RS message.

• Current default router has higher probability (=1) of responding to RS message.

• If current default router not present – a fast RA mechanism ensures arrival of a RA message from one of the router on the link.

Page 19: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Priority Landmark

Pros• No modification to RAs.• No router negotiations.

– Just listen to RAs.

• Can confirm bi-directional reachability of default router if no movement.

Cons• Relies on RA scheme

that ensures that first response is from requested router if present.

• IPR Considerations.• Adds router address

option to RSs.

Page 20: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Hybrid landmark

• Hybrid of requested and priority landmark.• Keep the possibility of another router providing

definite answer on link change.

Page 21: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Hybrid Landmark

Pros

• Simple RAs (yes/no flags).

• No router negotiations.– Just listen to RAs.

• Can confirm bi-directional reachability of default router if no movement.

• Allows any router to respond giving definitive answer to link-change question.

Cons• Best suited to unicast

RAs. Aggregation gets tricky.

• IPR Considerations.• Adds router address

option to RSs.

Page 22: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Getting the RA Quickly

Page 23: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Fast Router Discovery

• Cache recent RA messages in Access Points.• Access points forward RA message when hosts

associate with them.

Page 24: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Fast Router Discovery

Pros• Potentially very fast.• No changes to IP

routers or IP hosts.

Cons• Link specific.• Some security

concerns.• No changes to IP

routers or IP hosts – Why standardize? Why @ IETF?

Page 25: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Simple Fast RA

• Administratively configure one router to respond immediately.

Page 26: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Simple Fast RA

Pros• Simple.

Cons• Needs mechanism to

select fast router – administrative config.

• Single point of failure.• Unbalanced loading

of routers.• Will NOT work with

priority landmark (Is this a Con?).

Page 27: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Deterministic Fast RA

• Routers on a link negotiate amongst themselves an ordering for responding to solicitations.

• Responses are made in order at fixed intervals starting from zero delay for the first router.

Page 28: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Deterministic Fast RA

Pros• No inherent single

point of failure.• One lost RA may not

cause much of a slow down.

Cons• Relatively complex

negotiation.• Unbalanced loading

but fairly simple to change that.

• Security requirements among routers?

Page 29: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Negotiation-free Deterministic Fast RA

• Routers listen to other routers on link and create a token for each of them.

• XOR token with some ID of the received RS message (TSLLAO) and create ranking for response.

Page 30: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Negotiation-free Deterministic Fast RA

Pros• As for Det Fast RA• Routers just need to

listen to determine their delay for a given RS.

Cons• RS needs to include

some variable data like a TSLLAO.– May be able to use

source address if not :: and OptiDAD in use.

• Synchronized/secure knowledge of router tokens desirable.

Page 31: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Probabilistic Fast RA

• Listen for RA to learn number of routers on the link (bounded by MIN and MAX).

• When RS is received, calculate probability of response in particular slot based on the number of routers using pre-defined function (Loaded towards slot 0).

Page 32: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Probabilistic Fast RA

Pros• As for Det Fast RA• Routers just need to

listen to determine their delay.– No negotiation.

• No security required between routers.

Cons• Some responses will

be delayed (a small amount).

• IPR Considerations.

Page 33: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Putting things together (Examples)

Page 34: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Requested Prefix/Negotiation-free Deterministic FastRA

• Routers listen to RA messages and learn supported prefixes and IID of others routers – generate token based on the IID.

• Host include TSLLAO and a current prefix in the RS message.

• Router response ranking is created by XOR of TSLLAO with tokens of each router.

• Router responds with Yes/No flags to indicate the presence of the prefix identified in the RS in order of their ranking separated by pre-defined time-slots.

Page 35: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Priority Landmark/Probabilistic FastRA

• Routers listen to RA messages and learn number of routers on link. (Restricted by MIN and MAX).

• Hosts include current prefix and current router address in the RS message.

• If current router present in the link, will respond with RA message at slot = 0.

• Other routers respond at slot = [1..N] using a probability function of number of routers on link.

• Hosts will assume link change if first response is not from current router.

Page 36: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

CompleteRA /FastRA

• Routers listen to RA messages and learn supported prefixes of others routers.

• Host sends RS message.• Pre-configured router responds with complete list

of prefixes on link included in the RA message. • All other routers respond after random delay

between 0..500 ms, with complete list of prefixes in the message.

• Host assumes link change if its current prefix is not included in the complete prefix list.

Page 37: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Prefix LinkID/FRD

• Routers negotiate among themselves to identify one prefix as the LinkID for the link.

• Access points cache recent RA message(s).• Host attaches to a new Access Point.

– Access point transmit cached RA message to host.– Based on linkID host knows whether it has moved.

Page 38: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

AnalysisFiltering criteria

Page 39: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Generic Considerations

• Signalling overhead

• Interoperability with security schemes

• No additional security hole

• Interoperability with non-supporting nodes

Page 40: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Checking for Link Change

• Negotiated Vs non-negotiated.

• Security/Trust requirements between routers.

• Dependence on solicitation.

• Knowledge of all prefixes in all routers – Synchronization required.

• Multicast RA possible.

• Priority to current router.

Page 41: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Checking for Link Change (Contd.)

• Definitive answer from other routers (useful when there is packet loss?).

• Efficient when no change of link.

• Bi-directional reachability verification.

• Dependence on information on RS messages.

• Graceful recovery from changes in link configuration state.

Page 42: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Getting the RA quickly

• New signalling mechanism?

• Access point involvement?

• Negotiated vs non-negotiated?

• Security/trust requirements between routers?

• Delay?

• Dependence on information in RS messages?

Page 43: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Where to now?

Page 44: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Where to?

• Any schemes not yet considered?• Develop the filtering criteria.

– Levels of importance?

• Apply filtering criteria.– Around 35 combinations possible so far.– No obvious winning combination.– Need to prune the field down to a small number.

• Sort out the details.• Implementation bake-off?• Should the DT continue down this path?

Page 45: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Recent Developments

• Removing:– Link detection:

• Explicit LinkID – random• Explicit LinkID – prefix

hash• Priority Landmark

– Fast advertisement:• Simple FastRA• Deterministic FastRA

• This leaves:– Link detection:

• Explicit LinkID - agreed prefix

• CompleteRA• Requested Landmark• Hybrid Landmark

– Fast advertisement:• Fast Router Discovery• Hash-ordered FastRA• Probabilistic FastRA

• Some techniques probably not worth pursuing due to incompleteness or complexity of negotiation.

Page 46: DNA design team update JinHyeock Choi Tero Kauppinen James Kempf Sathya Narayanan Erik Nordmark Brett Pentland Design Team: Brett Pentland.

Thankyou.