Network Convergence
If a link or device failure exists,
the network needs to reconverge, providing alternate paths for data traffic.
Because physical devices are typically interconnected in somewhat of a ring
topology for redundancy reasons, deploying Layer 2 protocols that aid in
eliminating any logical loops is important
The three possible methods for Layer 2
loop avoidance are RTG, the Rapid Spanning Tree Protocol (RSTP), and the
Multiple Spanning Tree Protocol (MSTP).
The RTG feature eliminates the need to
configure STP on the switch. It is similar to RSTP root and alternate port,
without the need for configuring RSTP. Its ideal implementation is on a switch
with a dual-home connection, where one link becomes active and forwards
traffic, while the other link is blocking the traffic and is a backup to the
active link. It provides sub-second convergence and achieves loop prevention
without a spanning tree.
RSTP provides sub-second convergence
while avoiding STP slow network convergence. While STP can take 30 to 50
seconds to respond to a topology change, RSTP is typically able to respond to
changes within three times the Hello message interval. (The default value is 6
seconds.)
In the past, MSTP was defined in
Institute of Electrical and Electronics Engineers (IEEE) 802.1s, and it was
later added into the IEEE 802.1q standard. MSTP is an extension to RSTP and
further develops the creativity of VLANs. MSTP creates a separate spanning tree
for each VLAN group and blocks all paths but one of the possible alternate
paths within each spanning tree.
VSTP addresses the drawback of STP’s
and RSTP’s inability to utilize redundant paths to forward traffic, enabling
each VLAN to run its instance of the spanning tree. It is a non-standards based
protocol and it inter-operates with Cisco’s PVST
Because no dynamic path optimization
exists between the network tiers, a network failure can cause higher latency.
You should be aware of this drawback when designing networks for latency-sensitive
applications
Layer 3 at the Access Tier
Consider an alternative to the Layer 2
approach at the access tier. You can configure the access-core uplink as a
Layer 3 connection, which requires servers to have the network’s default
gateway pointing to the access switch. In this design, Layer 2 broadcast
domains can span across multiple member switches within the same access switch
Virtual Chassis. This setup implements multinode server cluster technology that
requires Layer 2 connectivity among the nodes participating in these clusters.
Some examples of these technologies include VMware live migration technology (VMotion),
Microsoft active clusters, and other high-performance computer clusters or
grid-computing applications.
The design requires the use of an
interior gateway protocol (IGP) — OSPF, for example. We recommend including equal-cost
multipath (ECMP) routing, which provides traffic load balancing between the
network tiers. The use of ECMP on the uplink LAGs, interconnecting the access
and core tiers, replaces STP and has many advantages
• Minimized Layer 2 broadcast domain,
• Ease of troubleshooting,
• Deterministic behavior for minimal
packet loss,
• Automatic load balancing at a
per-prefix or a per-packet level, and
• Dynamic network optimization and
path selection
This approach limits the Layer 2
domain to a single Virtual Chassis. It also limits Layer 2 mobility to a set of
access elements.
Network Convergence Across Access-Core Tiers
Because a single Layer 3 routing
domain directs network traffic flow, any failure in the link or network device
results in a topology change and network re-convergence across the entire
network. Network path optimization happens automatically, resulting in network
deterministic behavior, providing a consistent latency of traffic flow, and
allowing easier network provisioning.
