Implement a VXLAN-based network into an SoC

Ron DiGiuseppe, Synopsys
EDN (January 08, 2015)

With the growth in cloud computing applications, the industry is transitioning from traditional campus enterprise data centers to hyperscale cloud and mega data centers. According to IDC Research, server shipments to cloud service providers are expected to represent 43% of all servers by 2017[1].

Furthermore, the number of internet connected devices generating data traffic is expected to exceed 20 billion in 2018[2] driven by the increase in IoT applications. These industry trends put significant demands on all areas of large cloud data centers including compute servers, storage and network. To achieve higher performance and reduce the cost and time of deploying these new applications, cloud and mega data center operators are redesigning their data center networking architectures to utilize virtual environments. By using Virtual Extensible Local Network (VXLAN)-based overlays operating over 10G Ethernet IP, data center operators can simplify the management of the networks and eliminate network bottlenecks in hyperscale cloud computing data centers.

Virtualized networks have evolved from classic-tiered north-south access/aggregation architectures to flat leaf-spine topologies driven by the increasing amount of east-west server-to-server communication. However, the data traffic in flat leaf-spine network topologies is encountering bottlenecks due to limits in the network infrastructure. Since server-to-server data traffic can consume up to 80% of data communication within cloud data centers, the network must scale to meet the increasing demands. For example, the traditional Layer 2 (L2) networks defined in IEEE 802.1Q standard virtual local area networks (VLANs) tagging has a limit of 4094 IDs, which can easily be exceeded. New dense micro servers support up to 148 servers in 2-rack unit (2U) form factors and a top-of-rack (ToR) switch can connect hundreds of virtualized servers, each hosting up to 100 virtual machines (VMs), which stresses the network maximum limit of 4094 L2 MAC IDs.

Network overlay technologies such as VXLAN and network virtualization using generic routing encapsulation (NVGRE) solve the scalability limits of VLANs by “stretching” the L2 network. VXLAN is an L2 overlay scheme defined by the Internet Engineering Task Force (IETF) to provide a framework for overlaying virtualized L2 networks over Layer 3 (L3) networks[3]. The VXLAN framework is an IETF Informational Request for Comments (RFC), a document published for the general information of the Internet community by the IETF. Each overlay is a VXLAN segment where VMs within the same VXLAN segment can communicate with each other. VXLAN enables server virtualization by logically grouping the VMs residing on multiple locations over the L2/L3 network to form a virtual network segment that can have up to 16 million network segments. Each logical network is identified by a unique ID called a VNI of the VXLAN packet field. The VMs can communicate by switching between the network segments using software configuration via network management.

To read the full article, click here