Advances in Server Virtualization Provides Better Scalability and Redundancy

19th Dec 2014

Traditional data center networks using leaf-spine topology are based on Client –Server model with majority of the traffic from North to south bound. Recent technical advancements in Server Virtualization and Cloud Computing have improved the server resource usage by supporting multiple virtual machines per node. This resulted in new east – west traffic patterns making re-design of traditional data center networks to provide scalability and redundancy.

Advances in Server Virtualization Provides Better Scalability and Redundancy

In a typical leaf-spine topology, using xSTP to provide loop free topology resulted in under-utilization of redundant links between switches (xSTP blocks the redundant links) and higher convergence time when there is a node/link failure. TRILL and MLAG are considered as alternate solutions which provided better utilization of redundant links while overcoming the other shortcomings of xSTP.

TRILL provides optimal pair-wise forwarding, multi-pathing for both unicast and multicast traffic with quick convergence and loop-free path. While MLAG provides multi-homing support and node level redundancy with better utilization of the redundant links in the topology.

ZebOS TRILL Integration with MLAG

TRILL with MLAG solution can be used in a deployment scenario where TRILL is deployed in core to access layer, while using MLAG to provide multi homing functionality to the end servers supporting basic LACP.

In below mentioned deployment scenario, customer edge devices are multi-homed to rbridges. All of the up-links are considered as an MLAG interface. This active-active load-sharing mechanism will improve resiliency and maximize the available network bandwidth. TRILL between core to access layer helps to provide multipathing, optimal forwarding and quicker convergence.

ZebOS TRILL Integration with MLAG

Key design decisions taken for integration of TRILL and MLAG:

  • Traffic distribution mechanisms of TRILL Access node and MLAG
  • TRILL uses appointed forwarder mechanism which provides per vlan active-standby traffic distribution and loop avoidance in case of a multi-access link. Appointed forwarder Rbridge can ingress and egress the vlan traffic to/from the TRILL network. This requires edge Rbridges to exchange TRILL hellos on the access side, which carries enabled vlan information for appointed forwarder distribution.
  • As per IEEE 802.1AX-Rev standard, MLAG traffic distribution logic is based on conversation-id’s where admin can select different types of conversation id types.  Currently ZebOS MLAG supports VLAN based conversation id type and auto allocation of conversation-id’s between the MLAG portal systems based on the vlans configured.

When end host forms a multi-homed connection with multiple Edge TRILL RBridge using port-channel then TRILL hello’s sent from one Edge RBridge will not be received on other Edge RBridge. As a result every TRILL edge interface will consider itself as appointer forwarder for all the vlans. This leads to frame duplication, loop, address flip-flop and RPF check failure in TRILL cloud.

So TRILL appointed forwarder distribution logic is disabled on the access link and MLAG takes control of vlan distribution logic. TRILL will make use of the MLAG distributed vlan list for sharing its interested vlan’s to other Rbridges in the TRILL Cloud.

  • The TRILL switch ports attached to the MLAG in access link demarcate the edge of the TRILL and no adjacency can be formed on top of MLAG.

In case of Node or Link level failures at Access Layer, MLAG with fast timers provide faster convergence than TRILL.