Detour Problem in multi-homed Customer
Networks
NEC labs. America
Dirceu Cavendish
Kazuo Takagi
1
Assumptions based on current discussions
in IEEE802.1AD.
•
Assumptions:
1. Customer networks support legacy LAN/MAN/WAN
specifications only.
2. Customer network spanning trees are established
independently of those in the provider network.
3. Customer sites might be multi-homed to the provider
network.
4. Others…
2
Network assumptions
2. Establishing
independent spanning trees
1. Legacy LAN/MAN/WAN
1. Legacy LAN/MAN/WAN
ST for Customer A
RB
RB
Customer A LAN 1
Provider Network
RB
Customer A LAN 2
RB Root Bridge
3. Multi-homed connections
3
Provider Requirement : Loop prevention
• Current view
– Recognizing Provider network as “Fat Yellow Cable”.
• BPDUs from each customer port are broadcasted to all
customer ports attached, transparently.
As a result, each customer can construct (a) spanning tree(s) without incurring in loop.
Transparent transport of
customer BPDUs
ST for Customer A
Loop prevention network
RB
Customer A LAN 1
RB
Fat Yellow Cable
Customer A LAN 2
Provider Network
Blocking role ports
4
Customer Traffic Detour
• However, in this scheme, multi-home ports to customer
sites without a customer root bridge are all kept in forward
state”.
– Cost of links through the “fat yellow cable” look all the same.
Detour
Forwarding role ports
ST for Customer A
Region A
C:1
RB
C:1
C:10
RB
Cost(C): 10
C:1
Detour
C:10
Blocking role port
C:11
Region B
C:11
Fat Yellow Cable
Customer A LAN 1
C:11
C:10
Customer A LAN 2
MAC frames between Region A and B have to
be exchanged via provider network, even though
Region A and B belong to same LAN.
5
Detour Problems
• In general, bandwidth and traffic within customer LAN are
much larger than those in inter-LANs.
– Inefficient bandwidth usage for customers.
• Unnecessary customer traffic goes through provider network (waste
of customer money)
– Performance degradation of customer traffic (delay, loss) due to
long path and bandwidth bottleneck.
– Congestion spot more likely to appear within the provider network.
GbE network
Bottleneck
10Mbps
Region A
10Mbps
Region B
Detour access
6
Detour Avoidance
• Detour remains as long as:
– Multi-homed ports are in forward state.
• Objective: to avoid multi-homed ports in forwarding
state without new or complicated protocols
– Addition of Provider transport cost to Customer BPDU
• Port-cost modification in customer BPDUs at provider network.
– PEB/PBs calculate and overwrite the cost parameters in the customer
BPDUs.
7
Port-cost adjustment
• Each PEB/PE calculates and overwrites the costs
parameters in Customer BPDUs.
– It works for some instances:
Forwarding-role ports
Forwarding-role port
ST for Customer A
C:1
RB1 C:10
C:20
RB
C:1
C:40
C:30
C:50
C:40
C:40
C:51
C:51
C:30
C:1
C:50
C:51
C:50
C:60
– But not always: provider link cost may be such that path costs
through multi-homed interfaces are still the cheapest, resulting in
several access ports in forwarding state.
Forwarding-role port
Forwarding-role ports
ST for Customer A
C:1
RB1 C:10
C:20
C:1
RB
C:30
C:40
C:50
C:30
C:51
C:51
C:1
C:51
C:50
C:40
C:40
C:50
This scheme does NOT guarantee detour avoidance.
8
New Approach
• Objective:
– To avoid multi-homed ports in forwarding state without
complicated protocols:
• Approach
– Reuse legacy protocols.
• Use customer BPDUs (C-BPDUs)
– No complicated processes to both provider and
customer networks required.
• Provider and customers do not adjust the cost parameters in the
BPDUs.
• Customers do not need to be aware of provider network points
of attachment at all.
– No configuration procedure needed.
9
Possible Solution:
Single port selection via customer BPDUs (1)
•
Each PEB decides its port-role autonomously, via C-BPDUs sent
from both Provider and customer sides.
1. Root bridge IDs in C-BPDUs are used as LAN IDs.
2. Each PEB broadcasts C-BPDU to all the PEBs transparently, via the
provider spanning tree used for that customer.
3. Each PEB decides its port-role by comparisons of parameters, received
from C-BPDU with the same root bridge ID as the one PEB attaches to.
•
Parameters: cost, MAC addresses, ….
4. Each PEB discards C-BPDUs received from provider bridges.
Root:RB1
RB1
ST for Customer A
RB
For LAN 2
RB2
Customer A LAN 2
Customer A LAN 1
C-BPDU flows
For LAN 1
Root:RB2
Provider Network
C-BPDU processing
C-BPDU processing
10
Proposal:
One port selection with customer BPDUs (2)
•
Each PEB decides its port-role autonomously, referring C-BPDUs
sent from both of Provider and customer sides.
1. Root bridge IDs in C-BPDUs are used as LAN IDs.
2. Each PEB broadcasts C-BPDU to all the PEBs transparently, via the
customer spanning tree.
3. Each PEB decides its port-role by comparisons of parameters, received
from C-BPDU with the same root bridge ID as the one PEB attaches to.
•
Parameters: cost, MAC addresses, ….
4. Each PEB discards the C-BPDUs received from provider site.
Root:RB1
RB1
ST for Customer A
RB
For LAN 2
RB2
Customer A LAN 2
Customer A LAN 1
C-BPDU flows
For LAN 1
Root:RB2
Provider Network
C-BPDU processing
C-BPDU broadcast to PEBs
C-BPDU broadcast to PEBs
C-BPDU from LAN2
C-BPDU from LAN1
C-BPDU processing
11
Proposal:
One port selection with customer BPDUs (3)
•
Each PEB decides its port-role autonomously, referring C-BPDUs
sent from both of Provider and customer sides.
1. Root bridge IDs in C-BPDUs are used as LAN IDs.
2. Each PEB broadcasts C-BPDU to all the PEBs transparently, via the
customer spanning tree.
3. Each PEB decides its port-role by comparisons of parameters, received
from C-BPDU with the same root bridge ID as the one PEB attaches to.
•
Parameters: cost, MAC addresses, ….
4. Each PEB discards the C-BPDUs received from provider site.
Myport
role role
is
This
ST
for Customer A
isforwarding.
forwarding.
Root:RB1
RB1
Root:RB2
RB2
RB
Myport
rolerole
is
This
Blocking.
is Blocking.
X
X
Customer A LAN 2
Customer A LAN 1
C-BPDU flows
For LAN 1
For LAN 2
Provider Network
C-BPDU processing
C-BPDU broadcast to PEBs
C-BPDU broadcast to PEBs
C-BPDU processing
12
Proposal:
One port selection with customer BPDUs (4)
•
Each PEB decides its port-role autonomously, referring C-BPDUs sent
from both of Provider and customer sides.
1. Root bridge IDs in C-BPDUs are used as LAN IDs.
2. Each PEB broadcasts C-BPDU to all the PEBs transparently, via the customer
spanning tree.
3. Each PEB decides its port-role by comparisons of parameters, received from CBPDU with the same root bridge ID as the one PEB attaches to.
•
Parameters: cost, MAC addresses, ….
4. Each PEB discards C-BPDUs received from provider site – no BPDU
tunelling.
Root:RB1
Root:RB2
ST for Customer A
RB1
RB2
RB
X
X
Customer A LAN 2
Customer A LAN 1
C-BPDU flows
For LAN 1
For LAN 2
Provider Network
C-BPDU processing
Discarding
C-BPDU broadcast to PEBs
C-BPDU broadcast to PEBs
Discarding
C-BPDU processing
13
Customer/Provider Network
2. Establishing
independent spanning trees
1. Legacy LAN/MAN/WAN
1. Legacy LAN/MAN/WAN
ST for Customer A
RB
RB
RB
Customer A LAN 1
Provider Network
RB
Customer A LAN 2
RB Root Bridge
3. Multi-homed connections
14
This scheme merit:
Easy bridge role change(1)
Divide one LAN into two LANs
STEP 1
•
•
When a failure is occurred, TCN
is sent within LAN 2.
The attached PEB broadcasts
information with TCN to unlearn.
•
•
C-TCN
Forwarding or blocking
Failure
Customer A LAN 2
ST for Customer A
RB ID in C-BPDU is RB2
RB
X
RB2
ST for LAN 3
X
RB ID in C-BPDU is RB3
STEP 3
PEBs decide the bridge role.
ST for LAN 3
X
The PEB receiving TCN from
LAN is in blocking role, in
order to avoid loop.
New RB comes up in LAN.
PEBs broadcast information
with attached RB ID.
•
RB2
RB
STEP 2
•
•
ST for Customer A
Failure
RB3
New root bridge!
ST for Customer A
RB
RB2
ST for LAN 3
Failure
RB3
Forwarding role.
15
This scheme merit:
Easy bridge role change(2)
Connects two customer LANs
STEP 1
ST for Customer A
Back-to-Back
•
•
•
•
•
The PEB receiving TCN from
LAN is in blocking role, in order
to avoid loop.
TCN
RB3
STEP 2
ST for Customer A
RB ID in C-BPDU is RB2
RB
X
RB2
X
Only one RB comes up in LAN.
PEBs broadcast information
with attached RB ID.
RB ID in C-BPDU is RB2
STEP 3
•
RB2
RB
When two customer LANs are
connected back-to-back, TCN is
sent within LANs.
The attached PEB broadcasts
information with TCN to unlearn.
ST for Customer A
Forwarding role.
RB2
RB
PEBs decide the bridge role.
•
Forwarding or blocking
X
16
Merits of the proposal
• Efficient bandwidth usage
– Avoidance of detour
• Automatic port role decision without prenotification of LAN separation and integration.
• Not complicated scheme.
– Rely on C-BPDUs.
• C-BPDU broadcast has been already discussed for unlearning.
17
Proposal Requirements
• Each site is required to have a root bridge.
– This might avoid customers to make a single spanning
tree across the provider network.
• C-BPDUs are required to be broadcasted to all the
PEBs on the provider spanning tree for the
customers.
• Each PEB is required to discard the C-BPDUs
received from Provider side.
18
Summary
• The concept of “Fat Yellow Cable” is good for
avoidance of loop in Provider bridge.
• However, this has a detour access:
– In the sites without a root bridge, however, frames
might be transported via provider network, even when
transported to the terminal in the same LAN.
• This is because all or several attached ports probably has the
same cost value and they all are in forwarding role.
• Proposal:
– One port selection via customer BPDUs
• Each PEB decides its port-role autonomously, based on
C-BPDUs sent from both of Provider and customer sides.
19