1jf_topo_04.pdf802-17-01-00115

RPR Topology Discovery Proposal

Italo Busi, Alcatel

Jeanne De Jaegher, Alcatel

Jason Fan, Luminous

Chi-Ping Fu, NEC

John Lemon, Lantern

Robin Olsson, Vitesse

Harry Peng, Nortel

Yiming Yao, Luminous

2jf_topo_04.pdf802-17-01-00115

Components of a complete RPR Components of a complete RPR proposalproposal

Topology Discovery

Protection Switching

RPR MAC

MAC Service Definition

Media Access Control

Frame Format

Transit Path

Bandwidth Mgmt.

OAM and Layer Mgmt

Bridging

PHY interface

Ethernet

SONET/SDH

jf_topo_0x.pdf

jal_prot_0x.pdf

am_mrm_0x.pdf

hp_bwmgnt_0x.pdf

rb_phys_0x.pdf

hp_grs_0x.pdf

cb_layer_0x.pdf

hp_brcom_0x.pdf

rs_frame_0x.pdf

sa_transi_0x.pdf

ib_oamp_0x.pdf

3jf_topo_04.pdf802-17-01-00115

802.17 MAC Components

Reconciliation Sublayer(Ethernet/SONET)

Reconciliation Sublayer(Ethernet/SONET)

PHY (Ethernet/SONET)PHY (Ethernet/SONET)

MLME

RPR MAC Control

Topology

(Topology DB, State Machine, etc.)

Protection

(Steering DB, State Machine, etc.)

Ringlet Selection

MAC Client(s)MAC Client(s)

Rate Control / Policer

Transit Data Path

RPR Media Access Control

FCS

(PSAP)

(PSAP)

(MLSAP)

(MLSAP)

(MSAP)(RMSAP)

MAC Layer

Boundary

OAMP

4jf_topo_04.pdf802-17-01-00115

Information Sharing

• RPR Topology Image consists of station addresses, link connectivity between stations, and status for each link

• RPR Topology Image used by other algorithms– Steering algorithm uses Topology Image to know for which

destinations steering is needed– Congestion avoidance uses Topology Image to know where

congestion is being experienced– Ringlet selection uses Topology Image to know which ringlets are

available to reach each destination

5jf_topo_04.pdf802-17-01-00115

Goals of this Presentation

• In September the details of the topology discovery proposal were presented– Minor modifications made in the written proposal based on

discussions with a variety of companies

• This presentation will cover– Discussion of methods for topology discovery

– Additional simulation results

6jf_topo_04.pdf802-17-01-00115

Method 1: Neighbor Discovery and Distribution

• Each station:– Discovers neighbors using periodic messaging– Broadcasts neighbor information on change– Collects neighbor information from stations detecting change

– Verifies with adjacent stations that discovered topologies match– Updates topology database if verification successful

• On bring-up a station:– Follows the steps above with one modification

• Broadcasts neighbor information with version indication set to 0 to cause other stations to send neighbor information to it

7jf_topo_04.pdf802-17-01-00115

Method 2: “Chaining”

• Each station:– Sends a topology message on each ringlet periodically

• Each station adds its MAC address to each topology message– Collects topology information from each ringlet

– Needs to verify that information is received from each ringlet– Updates topology database if verification successful

• On bring-up a station:– Follows the steps above

8jf_topo_04.pdf802-17-01-00115

Goals Met by Both Methods

• Scalable from 1 to 100’s of stations• Determine/validate connectivity and ordering of stations on ring• Topology database updated based on dynamic addition and

removal of stations to/from the ring• Ensure all stations on the ring have a uniform and current image

of the topology• Immediate reaction to changes• Operate without any master station on the ring• Provide means of sharing additional information between

stations• Operate independently of and in the absence of any management

systems

9jf_topo_04.pdf802-17-01-00115

Additional Goals of Topology Discovery

• Robustness

• Minimize messaging overhead

• Discovery of spans on which there is a single missing link

• Usable with all supported topologies: multiple ringlet ring, linear (broken ring), and “star” (single station)

10jf_topo_04.pdf802-17-01-00115

Robustness (I)

• Methods for tolerance of message loss– Neighbor Discovery

• Send broadcasts on both directions on the ring• Validate discovered topology prior to updating stored topology

through version comparison with neighbors

– Chaining• Repeat topology discovery periodically• Validate discovered topology prior to updating stored topology

(?) or update stored topology after identical topology discovered multiple times consecutively

• Neighbor Discovery is not restricted to its current validation mechanism

11jf_topo_04.pdf802-17-01-00115

Robustness (II)

• Neighbor discovery uses broadcast, which has minimum interaction with the MAC as it is sent on the ring

• Chaining requires the MAC of every station to be able to update topology packets correctly for any station to discover the topology

A

B

C

D

EChainingMsg Path,Ringlet 1

ChainingMsg Path,Ringlet 0

12jf_topo_04.pdf802-17-01-00115

Minimize Messaging Overhead• M stations on ring• Neighbor Discovery is very efficient

– 1 broadcast per ringlet only on topology change, sent only by stations detecting a neighbor change

– 1 unicast per ringlet only on topology change, sent by each station directly to a new station

– Total of 3 broadcasts and M-1 unicasts per ringlet, only on topology change

• “Chaining” is less efficient– 1 “broadcast” per ringlet per station, sent every topology discovery

period– Total of M “broadcasts” per ringlet, every topology period

• Choice of mechanism for tolerance of message loss can come at the cost of efficiency

13jf_topo_04.pdf802-17-01-00115

Discovery of Spans with Single Missing Link

• Neighbor discovery enables discovery of single missing link– Bandwidth efficiency consideration for steering protection: rings

with steering can utilize spans with single available link

• “Chaining” does not enable discovery of single missing link– Bandwidth efficiency cost for steering protection– No bandwidth efficiency cost for wrapping protection

14jf_topo_04.pdf802-17-01-00115

Multiple Ringlet Support

• Neighbor Discovery enables discovery of more than two ringlets, and does not preclude RPR from supporting these scenarios

15jf_topo_04.pdf802-17-01-00115

Simulation Results (I)

• Set up– 256 stations– 200 km circumference– Dual ringlet ring

– 1 Gbps ring rate– Processing times for messages set to exponentially distributed

times around mean of 200 usec for Neighbor_Hello and 500 usecfor Topology_Status

– Neighbor_Hello_Timer: 0.5 seconds

– Topology_Stabilization_Timer: 2.5 seconds

16jf_topo_04.pdf802-17-01-00115

Simulation Results (II)

• Scenarios– (0) Bring up all 256 stations at once– (1) Starting with link 0->1 removed, then add it back

– (2) Two failures: links 0->1 and 127->128 are removed, then add 0->1 back

– (3) Three Failures: span 64-65 plus links in (2) are removed, then add span 64-65 back

• In all the scenarios above, the new topology (for adding link or span back) is discovered within 2.5 seconds.

17jf_topo_04.pdf802-17-01-00115

• Scenarios with lost Topology_Status messages– (4) Two messages lost in (1)– (5) Two messages lost in (2)

– (6) Four messages lost in (3)

• In scenarios (4)-(6), the new topology is discovered within 5 seconds.

Simulation Results (III)

18jf_topo_04.pdf802-17-01-00115

Topology Discovery Summary

• Many topology discovery objectives are met by both neighbor discovery and “chaining” methods

• Neighbor discovery method:– Is more efficient than “chaining” method– Can be designed to have lower failure probability than

“chaining” method, at cost of efficiency– Increases bandwidth availability in rings using steering

protection through usage of all available links– Has been simulated to show robustness to loss of

multiple messages– Has been simulated to show topology discovery in rings

with multiple link/span failures