trill spb-comparison-extract

© 2009 Cisco Systems, Inc. All rights reserved. Cisco ConfidentialTRILL Overview 1

TRILL and SPBOverview & Comparison

Ali SajassiPrinciple EngineerNSSTG CTO Group

August 14, 2009EDCS-805350


Agenda

� Quick Update on PBB & PBB-TE� TRILL Overview� SPB Overview� TRILL v.s. SPB Comparison� TRILL w/ MAC-in-MAC (E-TRILL)� OAM for E-TRILL


PBB Updates� draft-sajassi-l2vpn-vpls-pbb-interop.txt has been accepted

as IETF WG draft about six months ago- Discusses the use of PBB over VPLS to address service instance and MAC scalability issues- Discusses different interop scenarios between PBB and VPLS including gradual migration and the following scenarios• H-VPLS with 802.1ah Access Network• H-VPLS with Mixed 802.1ad and 802.1ah Access Network• H-VPLS with MPLS Access Network and 802.1ah u-PE• H-VPLS with MPLS Access Network and 802.1ah n-PE


PBB Update – Cont.� draft-sajassi-l2vpn-pbb-vpls-multicast.txt

- use of BGP for limiting the scope of broadcast per I-SID within a single VPLS instance (e.g., multicast pruning per I-SID)- haven’t done much w/ this draft

� draft-sajassi-l2vpn-pbb-vpls-cmac-flush.txt-To flush C-MAC addresses in a PBB-VPLS- C-MAC flushing is needed upon AC change – e.g., switch over from primary AC to backup AC- Independent from B-MAC flushing- Cisco started project IEEE 802.1Qbe which includes this draft & MIRP registration

� draft-mohan-l2vpn-vpls-oam.txt- Need to updated it and get it ready for WG draft call


PBB-TE� IEEE 802.1Qay has been completed

- Went through sponsor ballot a few meetings ago� It doesn’t seem like to have much traction in industry since

- the major SP sponsor of this technology switched gear to MPLS - the major vendor sponsor of this technology filed bankruptcy

� Instead there is lot of activity and vendor participation on MPLS-TP- Lots of draft submission and discussion on IETF MPLS WG - Lots of discussions between ITU & IETF groups on OAM mechanism for MPLS-TP- Lots of vendors have committed to implementing MPLS-TP including Cisco


Agenda

� Quick Update on PBB & PBB-TE� TRILL Overview� SPB Overview� TRILL v.s. SPB Comparison� TRILL w/ MAC-in-MAC (E-TRILL)� OAM for E-TRILL


On Collision Course

� Both IETF & IEEE have been working on next gen control plane for Ethernet � IETF TRILL currently targets Data Center & Enterprise segments� IEEE SPB targets Enterprise & Service Provider Segments� Each of the above technology can extend to the missing segment


Objectives for NG Ethernet Protocol� Optimum multicast & unicast forwarding� Fast Convergence� Robust loop mitigation and/or preventions� Scale for large networks� Ease of maintenance & reduce number of protocols� Scale to large number of MAC addresses – e.g., transparent to

the core nodes� ECMP !!� Multi-pathing for multicast traffic


Classical MSTP NetworkBridge DomainBridge DomainBridge Domain

CECECE

Cust. Eth Header

Cust IP Packet

802.1q Bridgeor HUB

Cust. Eth Header

Cust IP Packet

CECECE

CE1CE1CE1

CECECE

CE2CE2CE2

Links Disabled by Spanning Tree

Root


What is TRILL?� An IETF WG for Shortest Path Bridging� Control Plane leverages IS-IS, but is orthogonal to L3 IS-IS� Uses Routing Bridges (RBridges) using IS-IS to provide:

Shortest Unicast Paths (no STP single tree constraint)Faster Convergence timesMinimal or no configuration requiredLoad-splitting among multiple pathsLoop mitigation (TTL)*Support for multiple points of attachment

� Bonus: Relegates End Station MAC Address Learning to Edges, providing the MAC scalability


TRILL Basics� A TRILL Network is a collection of Routing Bridges, RBridges, or simply RBs� RBs may be interconnected by islands of 802.1 bridges

- RBs can be connected by multi-access links or - RBs can be connected by simple P2P links

� RBs use ISIS for discovery and to distribute Link State Databases� Packets are bridged between RBs, but Routed RBhop-by-RBhop from Ingress to Egress� Edge RBs learn End Station MAC addresses in the data plane and associate them with the edge RBs.


TRILL Network

Cust. Eth Header

Cust IP Packet

Cust. Eth Header

Cust IP Packet

TRILL Header

RBridge

802.1q Bridgeor HUB Next Hop(1)Eth Header

Cust. Eth Header

Cust IP Packet

TRILL Header

Next Hop(2)Eth Header

Cust. Eth Header

Cust IP Packet

CECECE

CECECE

CECECE

CECECE

CECECE


TRILL Basics – Multi-Access Link

� Support of multi-access link implies- Rbriges can replace IEEE bridges anywhere in the network- support gradual replacement of IEEE bridges by Rbridges- Existential threat to IEEE !!- mini-me IP encapsulation – e.g., outer MAC addresses have link local significance as opposed to 802.1ah which have system wide significance


TRILL Basics – Multi-Access Link - II

� Support of multi-access link introduces some complication to IS-IS procedures

-IMHO at least one-third of the TRILL spec is related to the support of multi-access link- Requires IS-IS to select a single DRB over the multi-access link- Requires that DRB to pick a designated VLAN- Requires that DRB to pick a appointed forwarder for each VLAN over the multi-access link- appointed forwarder can be same or different from the DRB


TRILL Routing� Unicast:

Packets are forwarded hop-by-hop using local Shortest pathEqual Cost Multipath supported, load-balancing a local decision

� Multicast:Traffic is sent over Multicast Distribution Trees (MDTs)Highest priority RB (Selector) decides how many MDTs to buildMDT list can be sent in LSP, but defaults to highest priority RBsIf # of MDTs < # of RBs, out-of-order packet delivery may occur*Building 1 tree similar to Spanning Tree except unicast still optimal

*when an unknown packet takes one path and a subsequent packet takes the known unicast path


TRILL NetworkRBridge DomainRBridge RBridge DomainDomain

MDT 3MDT 3MDT 3MDT 2MDT 2MDT 2

MTD 1MTD 1MTD 1 111

222

333

A TRILL Network and 3 MDTs rooted at different RBs


Loop Mitigation� There are three mechanisms for loop mitigation

- Color Blocking Logic (VLAN check)- Ingress Interface Check (source check)- TTL

� VLAN check avoid some frames from entering a transient loop; however, there can still be scenarios where frames can enter a transient loop� Ingress Interface Check prevents any frame from getting into a transient loop� TLL basically kills the frames already in a transient loop gradually – e.g., traffic in the loop decays as a function of time� When all the above three mechanisms are used together, then there would be no need for loop prevention mechanism – e.g., a solid loop mitigation mechanism removes the need for a loop prevention mechanism� TRILL uses all the above tree mechanisms together


TRILL Frame FormatOuter MAC DA

Outer MAC DAOuter MAC SA

Outer MAC SA

Eth = 802.1Q Outer VLANEth = TRILLEgress RBID Ingress RBID

V/M/R, Op, TTL

Inner MAC DAInner MAC DA

Inner MAC SAInner MAC SA

Eth = 802.1Q Inner VLANPayload ….

NextHop Address (of next RBridge)

E2EAddress

TRILLAddress

CHbH, CItE, Reserved…

Optional TRILLExtensions


RBridge Port Model

ISS (Internal Sublayer Service)

EISS (Enhanced ISS)

802.3 Phy

802.1/802.3 Low Level Control Frame Processing, Port/Link Control Logic

802.1Q Port Vlan Processing

RBridge High Level Control Frame Processing (BPDU, VRP)

RBridgeForwarding Engine, IS-IS, Etc.

Processing of Native and TRILL Frames

Existing 802 stds

New TRILL modules


TRILL Basics - Hellos� ISIS Hellos are sent using a MAC of All-IS-IS-RBridges� Hellos are used for neighbor discovery and exchange of info including:

RB System ID is 48 bits (typically MAC address)All standard ISIS infoDesired Designated VLAN, Designated VLANAnnouncing VLAN Set, Forwarding VLAN SetThe RBridge Nickname (to save space)

� RB Hello packets may be sent outside the RB network to discover and eliminate external loops


TRILL LSP� Contains:

ISIS IDs of neighbors + wide link metric (#22)A 16 bit nickname (negotiated on conflicts) and nickname priorityRB priority for choosing # of trees to calcNumber of MDTs to calculate (if this RB is highest priority)List of nicknames for root of Trees (if this RB is highest priority)List of VLANs for which this RB is appointed forwarder (shared access links only)


TRILL MGROUP-LSP� Brand new set of Multicast Group PDUs

- MGROUP-LSP (Link State PDU)- MGROUP-CSNP (Complete Sequence Number Packet)- MGROUP-PSNP (Partial Sequence Number Packet)

� Same procedures and format as Level 1 PDUS (LSP, CSNP, and PSNP)� GADDR TLV contains Group Address Sub-TLVS:

- GMAC-ADDR- GIP-ADDR- GIPV6-ADDR


Agenda

� TRILL Overview� SPB Overview (available at IEEE site)� TRILL v.s. SPB Comparison� TRILL w/ MAC-in-MAC (E-TRILL)� OAM for E-TRILL


802.1Q Data Plane Evolution

SA = Source MAC addressDA = Destination MAC addressVID = VLAN IDC-VID = Customer VIDS-VID = Service VIDI-SID = Service IDB-VID = Backbone VIDB-DA = Backbone DAB-SA = Backbone SA

2005 2008Standard Approved

1998

B-VIDI-TAG

B-TAG

B-DAB-SA

DASA

Payload

EthertypeC-VID

Payload

Ethertype

Q-TAG

DASA

C-VID

S-VID

Payload

Ethertype

C-TAG

S-TAG

DASA

I-SID

Payload

Ethertype

S-TAG

DASA

S-VIDC-TAGC-VID

Ethernet

ProviderBackboneBridges802.1ah

ProviderBridges802.1adEthernet

VLAN


Same Motivation as TRILL

• Traditional bridging based on RSTP/MSTP– Non-optimal forwarding– Manual configuration needed for disjoint trees & mapping of VLANs to these trees

Bridge DomainBridge Bridge DomainDomainCECECE

CECECE

CE1CE1CE1

CECECE

CE2CE2CE2

Root

CECECE

CECECE

CECECE

CECECE

CECECE

• Shortest path bridging– Optimum unicast & mcast forwarding

– Automatic SPT management controlled by IS-IS


IEEE 802.1aq variants� Shortest Path Backbone Bridging (SPBM) is aimed to be deployed in PBB networks

where all addresses are managed� Shortest Path Bridging (SPBV) is applicable in customer, enterprise or storage area

networks

SPBV SPBMMetro Core Network

• Reliability• Auto-discovery• Load sharing• Managed addresses

Access Network• Reliability• Bandwidth efficiency • Unknown or managedaddresses

Enterprise Network• Plug & Play• Easy to operate• Unknown addresses

MAC learningin data plane

MAC learningin control plane

SPB


SPT Region

MST Region

Interworking with RSTP and MSTP� Common Spanning Tree (CST)� Internal Spanning Tree (IST)� Common and Internal Spanning Tree

RSTP bridges

SPT Region

IST

MST Region

IST CST


IEEE 802.1aq uses IS-IS� Topology discovery

Each bridge is aware of the physical topology of the SPT Region� Service discovery

I-SID registrations are included into a new TLV� Shortest Path Tree computation� Maintenance of SPTs and CIST� SPTs can be set according to the discovered I-SID membership information

MRP is not needed� VID allocation to VLANs


Source tree identificationVLAN ID• An SPT is identified by the

SPVID assigned to the source bridge

� Applicable to both 802.1Q and 802.1ah bridges

� Ingress check on VID� Consumes VLAN space� Unidirectional VIDs

MAC address• B-SA and its Nickname incorporated into

Group MAC DA identifies an SPT

� Two VIDs only used for a whole set of Shortest Path Trees(Base VID and another VID)� Each ECMT uses one more VID

� Bidirectionality of VID is preserved� Only applicable to 802.1ah bridges� Ingress check on SA� All multicast addresses take the local bit

mapping


VID

VLAN assignmentVLAN Base VIDidentified by

MSTISPT Set

SPVIDBase VIDB-SA

Group B-DA

IST CST

Base VID Base VID

Learning

Base VIDID

LearningNon-learningPBB-TE

LearningNon-learning

SPBB

allocated to

supported bysupported by

Learning

IDID ID

MSTI CIST

Learning

SPT Set

VID*B-SA

Group B-DA

MSTI

Non-learningSPBB

IST

Base VID

Learning

IEEE 802.1aq implements

SPVID


Tree implementation• By Port Roles

• SPT is formed from the Root Bridge

• By Filtering Entries

• Source rooted SPTs applied in 802.1aq

Root Bridge Source Bridge

Designated,ForwardingRoot,ForwardingAlternate,Blocking


MAC learning� MAC learning in the data plane (Learning)

� MAC learning in the control plane (Non-learning)

Bridge

FDB

Bridge

FDB

Bridge

FDB

Frames

IS-IS IS-ISIS-IS

A d d r e s s e s A dd re ss es

Frames

A d d r e s s e s A dd re ss es

Bridge

FDB

IS-IS

Bridge

FDB

IS-IS

Bridge

FDB

IS-ISControl Messages Control MessagesAddresses Addresses Addresses


SPB� SPB (802.1Q compliant)

Uses VID for source identification, don’t own the C-MACSolution Attributes

VID Trees, one source per bridge, distributed in IS-ISSVL learning of unicast forwarding supported

Solution RequirementsMust Interwork at edges with RSTP, MSTPThe region may default to a single instance MSTP (associated with the “Base VID”) if the VID allocation fails or detects errorsMust support loop prevention, may support ingress check


SPB ConceptsSPT Region

SPVID = 45

SPVID = 41VID = 6

DASA

Payload45

DASA

Payload22

DASA

Payload

41

DASA

Payload22

DASA

Payload

22

CST IST

DASA

Payload45

DASA

Payload41

SPVID = 71

SPVID = 66 SPVID = 44

SPVID = 22

DASA

Payload

22

Base VID 22

SPB Region Identifier!= Base VID


SPBB� SPBB (Shortest Path Backbone Bridging)

Solution AttributesSingle VID for an SPT Region (may use VID Trees)Does not use learning of B-MACs

Provider addresses will all be known allows for more efficient flooding (no B-MAC broadcast storms),

Reduction in forwarding space Shared Forwarding, Solution Requirements

Must use Multicast loop Prevention, Must use ingress check for unicast


SPBB Operation

IS-IS IS-IS IS-IS

IS-ISIS-IS

IS-IS

BEB

BEBBEB

BCB BCBBEB “A”

BEB Backbone Edge Bridge BEB

PBBN

IS-IS

IS-IS IS-IS

Backbone Core Bridge BCB

Shortest Path Tree from “A”

Shortest path between any two points is both the same and symmetrical for unicast

and multicast


SPBB Shortest Path Tree to/from “A”

IS-IS IS-IS IS-IS

IS-ISIS-IS

IS-IS

BEB

BEBBEB

BCB BCBBEB “A”


PBBN

IS-IS

IS-IS IS-IS


Uses the full mesh network

Shortest path between any two points is both the same and symmetrical for unicast

and multicast

All pairs shortest path computation

performed in parallel


SPBB Multicast Groups

IS-IS IS-IS IS-IS

IS-ISIS-IS

IS-IS

BEB

BEBBEB

BCB BCBBEB “A”


PBBN

IS-IS

IS-IS IS-IS


I-SID 5

I-SID 5I-SID 5

MMAC for 5 from A

I-SIDs define efficient subsets


Forward and Reverse path Congruency

� Necessary if MAC learning is in the data plane� Not necessary if MAC learning is in the control plane� Going to be assured by both SPB and SPBB

Bridge5

11

1

1

1

1

2

Bridge2

Bridge1

Bridge3

Bridge4

Bridge6


Unicast and MulticastCongruency

� Necessary for MAC learning in data plane� Necessary for the proper operation of 802.1ag E-OAM� Going to be assured by both SPBV and SPBM

Bridge5

11

1

1

1

1

2

Bridge2

Bridge1

Bridge3

Bridge4

Bridge6

unicast

multicast


Implementation ofCongruency� Tie-breaking extension to Dijkstra for the case of equal cost multiple paths

List of node IDs comprising a path are unique{1,6,5} < {1,2,3,5} < {1,2,4,5}

� Same algorithm is used both for unicast and multicast

Bridge5

11

1

1

1

1

2

Bridge2

Bridge1

Bridge3

Bridge4

Bridge6


Load sharing� Two trees are calculated taking advantage of equal cost multiple paths:

{1,6,5} < {1,2,3,5} < {1,2,4,5}� SPT Primary Set � Primary Base VID� SPT Alternate Set � Secondary Base VID

Bridge5

11

1

1

1

1

2

Bridge2

Bridge1

Bridge3

Bridge4

Bridge6


Loop Prevention and Mitigation� Inconsistent view on network topology at different nodes may cause transient loops

in case of a link-state control protocol � Loop prevention

Tree Agreement Protocol (TAP)Handshake mechanism between neighborsExtension to MSTP’s handshake

� Loop mitigationIngress Checking (e.g. RPFC)Frames not arriving on the shortest path from the Source Bridge are discardedMakes the tree directedGood for loop prevention in most casesTransient loops may appear

Severe problem for multicast trafficA chance of network melt-down remains if one does not care

Ingress filtering has to be modified


Neighbor handshake mechanism� Let’s make it sure that bridges having different view on network topology do

not exchange frames� The link between adjacent neighbors has to be blocked after a topology

change until they agree that both of them have the same topology database� The agreement between neighbors is implemented by a handshake

mechanism� A digest of the topology database is exchanged

CRCCryptographic hash function (e.g. SHA-256)

� Agreements at different part of the network are independent of each other


Handshake: MSTP extension

• Tree Agreement Protocol (TAP)

• Two-way Agreement = three-way handshake

• No per tree handshake• BPDUs contain

– Digest of LSP database– Info on the CIST

• Proposal-Agreement– Explicit on the CIST– Computed for SPTs

Prop

osal

Agree

ment

Agreement


Handshake: Filtering entry manipulations

• SPBB networks• STPs are implemented by

Filtering Entries• Do not implement the TAP

extension to MSTP• Implement link-state database

synchronization (TAP logic)• Loops for unicast flows are

mitigated by Ingress Checking (RPFC)

• Remove ‘unsafe’ entries if neighbors are unsynchronized

Wait for LSP update

Unicast computation

Install Unicast andremove ‘unsafe’

Multicast FDB entriesUpdate Digest andsend it to neighbors

Multicast computation

Install ‘safe’Multicast FDB entries

Wait for Digest synch

Install ‘unsafe’Multicast FDB entries


Agenda

� TRILL Overview� SPB Overview� TRILL v.s. SPB Comparison� TRILL w/ MAC-in-MAC (E-TRILL)� OAM for E-TRILL


Functional ComparisonTRILL 802.1aq

Multipoint L2 Services: E-LINE, E-TREE, E-LAN

Yes Yes

C-MACs are learned only at the edges

Yes Yes

Optimum Unicast Fwding Yes YesOptimum Mcast Fwding Yes YesMulti-homing Yes

(using IS-IS)Yes

(using CST)ECMP Yes NoMulti-pathing via multiple trees

Yes Yes


Functional Comparison – Cont.TRILL 802.1aq

Ease of Provisioning Yes Yes (SPBV) No (SPBM)

Enterprise/DC Applicability Yes YesSP Applicability No YesCongruency unicast & mcast

No Yes

Congruency forward & reverse (not needed)

No Yes

Intra-region loop prevention

Yes Yes

Inter-region loop prevention

No (note-1) Yes (via CIST)

Note-1: but inter-region of .1aq is consider as intra region of TRILL


Control PlaneTRILL 802.1aq

Neighbor and topology discovery using IS-IS

Yes Yes

VLAN pruning using IS-IS Yes YesMulticast Pruning using IS-IS Yes YesDual-homing using IS-IS Yes NoNode address distribution in IS-IS (Nickname v.s. B-MAC)

Yes Yes

P2MP MDT setup using IS-IS Yes YesMP2MP MDT setup using IS-IS Yes No

Yes


Data PlaneTRILL 802.1aq

Multi-Access link support Yes NoLink-local outer MAC Yes No

Hop-by-hop forwarding Yes YesP2MP MDT Yes YesMP2MP MDT Yes No

(cannot do MSTP func.)TTL support Yes No

RPFC support Yes Yes


Functional ComparisonTRILL 802.1aq

Load Balancing at the edge nodePer-L2 flow Yes noPer-L3 flow Yes noPer-L4 flow Yes noLoad Balancing at the core nodePer-L2 flow Yes noPer-L3 flow Yes noPer-L4 flow Yes no


Scalability ComparisonTRILL 802.1aq

MAC scalability Yes YesService Instance Scalability

4K 16M

Multi-pathing for unicast Yes (Note-1) NoMulti-pathing for mcast Yes (Note-2) No

Note-1: only a single tree per node is requiredNote-2: supports MP2MP MDT which means for majority of network configuration only a few MDT needs to be built


Loop Mitigation & PreventionTRILL 802.1aq

Loop Mitigation: VLAN color blocking

Yes Yes

Loop Mitigation: RPF check

Yes Yes

Loop Mitigation: TTL Yes NoLoop Prevention: Handshake

No Yes


Agenda

� TRILL Overview� SPB Overview� TRILL v.s. SPB Comparison� TRILL w/ MAC-in-MAC – E-TRILL� OAM for E-TRILL


Extended TRILL� Two major short comings of TRILL that makes it

inadequate for SP space are:- Service Instance Scalability – need lot more than 4K- OAM support

� E-TRILL is intended to address these two issues and retain all the features & advantages of TRILL

- Use 802.1ah encap w/ TTL added to I-tag- Use TRILL as IS-IS control plane instead of 802.1aq


Service Provider Use Case

Cust. Eth Header

Cust IP Packet

Cust. Eth Header

Cust IP Packet

802.1ad Header 802.1Q HeaderCust. Eth Header

Cust IP Packet

Ext TRILL Header802.1Q Header

BEBBEBBEB

MetroEMetroEMetroE

BEBBEBBEB

BEBBEBBEB

BEBBEBBEB

BCBBCBBCB

BCBBCBBCB BCBBCBBCB

BCBBCBBCB

BCBBCBBCB

BCBBCBBCBBEBBEBBEB

BEBBEBBEB

BEBBEBBEB

Ext TRILL Header

802.1ad Network802.1ad 802.1ad NetworkNetwork


Extended TRILL� Leverage TRILL control plane and some of its

forwarding techniques:– Support for ECMP is a big win-win– TTL is essential for solid loop mitigation– Support for dual-homing on the access side– If congruency is needed, then it can be supported; however,

• Forward/reverse congruency is not needed because there is no MAC learning in data plane• mcast/unicast congruency is not needed if new OAM for TRILL is devised

� 802.1ah adds advantages of its own– BMACs allow collapsing Outer MAC header + TRILL Header– I-SIDs scale to millions of services instead of just 4K


Eth = .1ahI-SID C MAC DA

PCP/R, I-SID

Current Frame FormatsOuter MAC DA


Outer MAC SA

Eth = 0x88a8 B VLAN

C MAC DAC MAC SA

C MAC SA Eth = 802.1QC VLAN

Payload ….Payload

802.1ahOuter MAC DA


Outer MAC SA

Eth = 802.1Q Outer VLANEth = TRILLEgress RBID Ingress RBID

V/M/R, TTL

Inner MAC DAInner MAC DA

Inner MAC SAInner MAC SA

Eth = 802.1Q Inner VLANPayload ….

TRILL

NextHopAddress

E2EAddress

TRILLAddress


Proposed P2P frame format

Eth = PR-TAGSID C MAC DA

PCP/TTL, SID

Outer MAC DAOuter MAC DA

Outer MAC SAOuter MAC SA

Eth = 0x88a8 Outer VLAN

C MAC DAC MAC SA

C MAC SA Eth = 802.1QC VLAN

Payload ….Payload

E-TRILL � Outer MAC DA is encoded as:

– Unicast: OUI + 16b eRBID– Multicast: 802.1ah format for multicast (RootRB +ISID)

� Outer MAC SA is encoded as OUI + 16b iRBID� PR-TAG is a variant of the

I-TAG with a TTL and 20-bit I-SID