What is Virtual Chassis Technology? Virtual Chassis (VC) is a network virtualization technology offered in several models of Juniper Ethernet switches, such as the EX4200. With VC, between 2 and 10 physical Ethernet switches can be “stacked” to form a single logical form factor with a unified control plane and configuration file, as well as a single OS instance that operates across the entire stack. Route and switch engine redundancy are provided by the VC master and the backup through the creation and maintenance of synchronized forwarding tables and the exchange of stateful protocol information. The VC can span anywhere from 5 meters in height and breadth to 50 km across an entire campus or regional area! Using the EX4200 as an example, between 24 and 480 10/100/1000 Mbps ports can be managed as if they belong to one VC that supports multiple VLANs, a non-blocking architecture and full cross-sectional bandwidth utilization. Connecting the members of a Virtual Chassis in a ring topology results in a total virtual backplane speed of 128Gbps– 64Gbps in each direction. A key factor to understand is that Virtual Chassis technology deploys a direct path algorithm known as the Virtual Chassis Control Protocol (VCCP) for optimal traffic flow. Forwarding tables are built with the aggregate intelligence of all the members of the VC, thereby ensuring that traffic takes the shortest path to its destination. VC is one of the core enabling technologies of Juniper’s vision for the data center of the future, otherwise known as Stratus. The foundational concept behind Stratus is to replicate the simplicity of a single fabric in a single switch chassis across the data center – or between data centers! The result is a resilient low-latency data center fabric that has the built in flexibility to support converged networks and virtualized workloads in a variety of different topologies.. Building a Virtual Chassis As noted above, the physical members of a VC can be collocated in a configuration that spans no more than 5 meters between stack elements. A VC configuration consisting of adjacent switches interconnected with special 0.5 meter, 1.5 meter, 3 meter, or 5 meter VC port cables is called a dedicated configuration. A VC configuration interconnected via Gigabit Ethernet or 10-Gigabit Ethernet uplink ports across distances up to 50 km is called an extended configuration. Figure 2 shows a rear view of a Juniper EX4200 Ethernet switch and the VC Ports (VCP) that interconnect the different VC elements. Each VCP offers 64Gbs of throughput for data traffic as well as protocol state information to maintain synchronized control and data planes between the Master and the Backup Route-Switch Engines (RSE). Once the VCP ports are cabled, they are automatically enabled to provide the switching backplane for the dedicated chassis configuration. To extend the VC configuration across a campus or regional area, the EX-UM-2X4SFP 10Gbps module, with dual 10G ports, can be deployed and configured to carry VC control information over an aggregated link that provides up to 40Gbps of full duplex throughput – 20Gbps in each direction. Virtual Chassis Member Roles – Dedicated ChassisEach member in a Virtual Chassis configuration is assigned a specific role. A role determines the functions the member performs in the configuration. One member is assigned the master role and is responsible for managing other members in the Virtual Chassis configuration. Another member is assigned the backup role and takes over the master role if the master switch fails. All other members are assigned the line card role. The system executes a mastership election algorithm to determine member roles.Virtual Chassis Roles – Extended ConfigurationIn an extended configuration, the master and the backup should be in different locations as part of dedicated chassis configurations. Moreover, the extended VCP connections can consist of gigabit, ten gigabit or aggregated Ethernet port connections for higher bandwidth requirements.Cabling OptionsThere are three different methods for connecting members of a Virtual Chassis. These are the daisy-chained ring, the braided-ring and the extended configuration. The daisy-chained and braided ring methods are deployed in dedicated chassis configurations, with the braided-ring offering a greater maximum distance between the first and last member of the stack. See figure 6a and 6b below.As noted earlier, an extended chassis consists of smaller dedicated virtual chassis interconnected by Ethernet ports in a singular or bundled fashion. This capability provides the flexibility to manage an entire campus LAN as one management plane with one IP address and a singular control plane that allows for automatic configuration and code upgrades upon introduction of a new member to the VC. This is known as dynamic installation. See figure 6C below. Virtual Chassis Deployment OptionsData Center Top of Rack (ToR)The data center top of rack deployment is suitable for environments in which members of the Virtual Chassis configuration are collocated with servers in the same rack. A single Virtual Chassis configuration consisting of multiple switches in the same rack simplifies management by reducing the number of managed devices. This deployment also provides servers with the capability of configuring NIC teaming (LAG) to multiple members of the same Virtual Chassis configuration, increasing the total server network bandwidth while providing server link redundancy. See figure 7 below. As figure 8 shows, in a Dell m1000e blade server deployment that includes M-series stackable blade switches, virtualization of the edge and access layer can be achieved, thereby removing the need for the deployment of the Spanning Tree Protocol. Multiple physical connections between the blade switches and the ToR VC will be seen as one logical link by each virtual layer. This allows for a non-blocking rack architecture and full utilization of available cross-sectional bandwidth. Single Virtual Chassis Spanning Entire RowDepending on server density and spacing the ToR solution can be extended to support an entire row of cabinets – 10 cabinets if one switch is placed at the top of the rack or 5 if the design calls for a self-contained modular cabinet that can be removed or deployed at will. In the latter, all the server-to-ToR cables will be confined to the cabinet. The ToR switches shown below all belong to the same Virtual Chassis. End of Row (EoR)Because the EX4200 Virtual Chassis can support layer 2 and layer 3 deployments, it can also play the role of an end of row solution. Typically, the EoR is a layer 3 aggregation/distribution layer that includes the deployment of transparent data center services, such as load balancing, firewalls, SSL offloads, etc. Campus LAN Wiring Closet(s) and Campus AggregationHopefully by now, the concept should be clear: the Virtual Chassis technology allows the network architect to be extremely flexible and creative in deploying design solutions. A VC can be deployed in many different topological scenarios and in different network roles. In a campus environment, a VC can be formed in a single wiring closet or in several closets across a user-floor. The access layer and the campus distribution layer can both exist as VC instances. Wiring closets that are more than 5 meters apart can be connected via gigabit, 10-gigabit and aggregated Ethernet connections, just as they were in our data center examples. The campus aggregation layer can consist of a single Virtual Chassis instance or multiples instances, depending on the traffic requirements. The thing to keep in mind is that deploying separate VC instances in the aggregation layer will require the deployment of STP between VCs and the access layer, thereby mitigating the benefit of virtualizing network layers.See figure 11 below.SummaryDeploying a Virtual Chassis solution in a high performance data center reduces the complexity of switch hardware installations, OS upgrades, configuration tasks, and management of the switched fabric. Moreover, virtualizing multiple layers is one of the necessary steps toward the creation of a “flat” data center network in which all L2 links are utilized and multiple layers of packet processing are eliminated. This will minimize latency, remove the need to deploy L2 loop mitigation technology, and create a non-blocking fabric with an average X2 increase in available cross-sectional bandwidth across the data center.