Transcript egvrpv3_15.ppt

EGVRP/GVRP Simulation
IEEE 802.1 May 2004
Guyves Achtari
Paul Bottorff
EGVRP Basic Concepts
• S-VLAN Distribution Protocol for Provider Bridges
• Supports large S-VLAN address spaces up to 224
• Uses a network wide address size parameter to
determine the number of active bits from 12 to 24
• Maintains hard state to scale better
• Supports a “Dense” protocol mode for startup
• Supports a “Sparse” protocol mode for add/change
updates
• Normal carrier operation would only use “spare”
mode
EGVRP’s dense mode
index
range
4095 16773120 … 16777217 2^^24
…….
224 S-VLANs require 4095
fully populated frames of
4096 dense packed
attributes to update the
entire database.
5 20482 … 24577
2^^15
4 16386 … 20481
2^^15
3 12290 … 16385
2^^14
2
8195 …. 12289
2^^14
1
4098 …. 8193
2^^13
0
1
2^^12
….. 4097
MIB: Unified Address Size
header
type =2 index
4096 encoded vlans
dense mode
Each 4096 S-VLANs are grouped together. Each
group is represented by an array of 4096 state
machines. Up to 4096 indexed arrays correspond to
224 S-VLANs.
index:4095
One Applicant
engine per index
index:0
Applicant engine: Registrar engine:
1 global
1 leave timer
Transmit PDU
per vlan
EGVRP model
index:4095
One Applicant
index:4095
One Applicant
engine per index
index:0
Applicant
engine:
ONE global
Transmit timer
engine per index
index:0
Registrar
engine:
1 leave timer
per VLAN
Applicant
engine:
ONE global
Transmit timer
for all VLANs
GIP
for all VLANs
Registrar
engine:
1 leave timer
per VLAN
MAC Relay Entity
header
type =2 index
header
type =1
4096 encoded vlans
Attribute List
dense mode
sparse mode
Leave timers are internal
timers. They do not regulate
transmissions. Only one
Transmit timer per index per
port regulates all transmissions
for a set of S-VLANs
EGVRP/GVRP Simulator
• Written in ‘C’ code
• Allows creation of any bridge topology
• Creates a model with asynchronous operation of
each bridge modeled in the topology
• Simulates EGVRP and GVRP state machines
• Simulator is new and still under test. All results
are preliminary and still being verified.
Bridge Node of Simulation Model
Propagation
delay
Port 0
Port 1
Port 2
State machine
update delay (per vlan)
State machine
update delay (per vlan)
State machine
update delay (per vlan)
Code Calculation delay
( dense mode, per vlan)
Code Calculation delay
( dense mode, per vlan)
Code Calculation delay
( dense mode, per vlan)
Pack/unpack
delay
Pack/unpack
delay
Pack/unpack
delay
link
delay
link
delay
link
delay
Simulated Bridge Assumptions
• Bridge’s control plane can process 10K EGVRP
frames/second
– Order of magnitude faster than today’s Bridge control
planes
– Simulation times are normalized to GVRP rates
• Bridge’s S-VLAN database can be updated fast
enough to keep up with the EGVRP frames/sec
processing rate
– Allows including the database update times into the
packet processing time
Network Under Simulation
19 nodes
-Simulation of a tree
topology with 19 nodes
-The 10 edge nodes contain static initial S-VLAN
databases
- Each initial S-VLAN database is different from all
other initial databases
-All S-VLANs are configured in at least 2 edge
nodes
- Convergence is achieved when all S-VLAN
databases in all nodes are identical
Preliminary Simulation Result
EGVRP With 212 to 223 S-VLANs
Convergence
Time
Normalized
To GVRP
For
212-2 VLANs
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
EGVRP
GVRP
212-2
0
2000000
~221
4000000
6000000
8000000 10000000 12000000 14000000
S-VLANs
Simulation shows convergence time is almost flat for
increasing S-VLANs address spaces
Future Work
• Perform simulations which vary the relationship
between the protocol processing time and database
update time for large S-VLAN spaces.
• It is believed the simulations results are heavily
dominated by the 200 msec transmit timer. We will
investigate alternate timer values to determine the
impact on the convergence time.
• Perform simulations over more varied topologies
including rings of trees and larger populations up to
100 node networks.
• Investigate the relative performance of EGVRP to
GVRP with large S-VLAN spaces.
• Further simulations will be done to determine if dense
mode EGVRP really provides a significant
performance advantage.
Backup
Simulation model for protocol delays (dense mode)
Maximum
protocol cost at
start-up, when
each sub-set of a
set provokes two
join declarations
In compact mode 4096 S-VLANs
are packed in a frame.
Worst case: happens when sub-sets
of a set of S-VLANs can not merge
their declarations before the
transmit timer for that set expires
Example below: If different subsets of the same set reach a bridge
while the timer for that set has not
expired, declarations can be
merged and sent together
4096 encoded vlans
header
type =2
1
1
header
type =2
1
1
20-40
4097
40-60
4097
leaf
range
1
20-40
2
40-60
case 1: see explanation in notes
time
case 2: see explanation in notes
time
Transmit timer
time
case 3: see explanation in notes
leaf1
leaf2
Tree topology: same load, expanded
network
9 nodes
Node specifications (Same as before)
19 nodes
39 nodes
Simulation results:
This chart shows convergence time in a tree topology network with 9,19 and
39 nodes
time
10000000
1000000
100000
9
19
39
nodes