Switched Layer 2 Ethernet cores are
very cost-effective, and they provide high-performance connectivity
between switch blocks. The classic design model has several switch
blocks, each supporting Layer 2 devices in a wiring closet that
terminate into a Layer 3 device. A core composed of Layer 2 devices
connects
the Layer 3 devices.
The Spanning-Tree Protocol represents
a practical limit to the scale of a Layer 2 switched backbone. As
you increase the number of core devices, you need to increase the
number of links from the distribution switches to maintain
redundancy. Because routing protocols dictate the number of
equal-cost paths, the number of independent core switches is
limited. Interconnecting the core switches creates bridging loops.
Introducing the Spanning-Tree Protocol into the core compromises the
high-performance connectivity between switch blocks. Ideally, Layer
2 switched backbones consist of two switches with no Spanning-Tree
loops in the topology.
In Figure ,
the Layer 2 switched backbone provides redundancy without any
Spanning-Tree loops. Because the two core switches are not linked
together, loops do not occur. Each distribution switch in every
switch block has a separate path to each core switch. The dual
connection between the core and distribution device provides each
Layer 3 device with two equal-cost paths to every other router in
the campus.
Most of the designs successfully
follow the Layer 2/Layer 3/Layer 2 model. However, there are designs
where the Layer 2/Layer 3/Layer 3 model is advantageous.
You would implement a Layer 3 core,
as shown in Figure ,
for the following reasons:
- Fast convergence
- Automatic load balancing
- Elimination of peering problems
Fast convergence
As you increase the number of switch blocks and servers, each
distribution-layer device must be connected to the core. Because
there is a limit to the number of switch blocks in a dual Layer 2
core, increasing the number of connections means increasing the
number of core devices. To maintain redundancy, the core devices
must be connected. When you interconnect Layer 2 devices, bridging
loops appear. To eliminate bridging loops in the core, you must
enable the Spanning -Tree Protocol, which has a convergence time of
over 50 seconds. If the network core has a fault, Spanning-Tree
Protocol convergence can disable the core for more than one minute.
If Layer 3 devices are implemented in
the core, the Spanning-Tree Protocol is not necessary. In this
design, routing protocols are used to maintain the network topology.
Convergence for routing protocols takes from 5 to 10 seconds,
depending on the routing protocol.
Automatic load balancing
Load balancing allows for a traffic-distribution pattern that best
utilizes the multiple links that are providing redundancy. With
multiple, interconnected Layer 2 devices in the core, you must
selectively choose the route for utilizing more than one path. You
then manually configure the links to support specific VLAN traffic.
With Layer 3 devices in the core, routing protocols can be used to
load balance over multiple equal-cost paths.
Elimination of peering problems
Another issue with the Layer 2 core in a very large network is that
of router peering. Router peering mandates that the routing protocol
running within a router maintains state and reachability information
for neighbor routers. In this scenario, each distribution device
becomes a peer with every other distribution device in the network.
Scalability becomes an issue in the configuration because each
distribution device has to maintain state for all other distribution
devices.
By implementing Layer 3 devices in
the core, a hierarchy is created, and the distribution device is not
considered a peer with all other distribution devices. This type of
core may appear in very large campus networks where the network
supports more than 100 switch blocks.
Performance and cost issues Implementing Layer 3 devices
in the core is expensive. As stated earlier, the main purpose of the
core is to move packets as quickly and efficiently as possible.
Although Layer 3 devices can support switching of some protocols,
both performance and equipment costs weigh in heavily as design
considerations.
|