A few years ago, virtualization and cloud technologies have become ever more popular, and enterprises’ perceptions of server resource consumption (including CPU, RAM, storage, etc.) has increased dramatically. It results in the demand for the virtualized data center. In virtualized data centers, each physical server can run one or more virtual servers based on hypervisor platform.
Network architecture now faces the need for interconnection between virtual servers, on the other hand, the movement of virtual servers from one physical server to another must be quick and easy. In order to meet these requirements, various overlapping techniques have been developed, including NVGRE, SPBV, TRILL, Fabric Path, VXLAN… This paper will cover the most common fabric techniques Path, TRILL, and VXLAN.
Virtualization of physical servers, especially in large data centers, requires both large-scale distributed application support and the flexibility to move virtual servers among different locations in the data center. This requires the need to develop a second layer encapsulation technique that enables communication between servers to be flexibly expanded. For example, when extending the virtual server system between physical servers in different datacenters, these virtual servers must be connected together quickly, simply and visually as they are in the same LAN segment. Cisco has developed the Fabric Path to meet these new requirements. Fabric Path provides a layer 2 fabric with great expansibility.
While retaining the plug-n-play feature in the classic Ethernet, Cisco FabricPath provides high-capacity, high-anticipated, forwarding optimal packet capabilities by using a routing protocol SPF is the basis (IS-IS) for adapting to the challenge of processing small-size MAC address tables in large layer 2 domains (between data centers).
For frame forwarding, FabricPath uses an internally assigned Mac address hierarchy. FabricPath encapsulates the original Layer 2 frame with a new source and destination MAC address, FabricPath tag, and a new CRC. The FabricPath network uses new MAC addresses containing 12 identifier bits called SwitchID to forward packets to the next hop in the core network.
TRILL is a technology that meets the same requirements as Cisco FabricPath and has the same advantages as FabricPath. Unlike FabricPath, TRILL is an IETF standard proposed to address traditional STP challenges, including ineffective link utilization, slow convergence, and MAC address table expansion in the data center network.
Like FabricPath, TRILL uses IS-IS as a control protocol with the idea of utilizing Layer 3 routing protocols and simultaneously maintaining a simple Layer 2 network for packet forwarding in the data center.
For frame forwarding, TRILL uses the MAC-in-MAC encapsulation format. Rbridge encapsulates the original Layer 2 packet with a new destination MAC address (corresponding to the MAC address of the source Rbridge and the next-hop Rbridge), the TRILL header and the new CRC. Tag 802.1q or q-in-q should be stored in the internal header. The output rage removes the new header and frames forward based on the original frame information.
Today VXLAN has become the most common overlay protocol. VXLAN supports many vendors, including Cisco, VMWare along with the fact that it runs through the core IP network, which makes VXLAN a popular choice for deployment in the hub. large-scale data.
VXLAN uses VXLAN-VTEP tunneling devices to map endpoints to network segments. We can imagine that VTEP has two interfaces – one connected to the middle IP network and the other one connected to the internal network segment – VXLAN packaged support.
At present, many virtual switches have support VXLAN packaging. Data center customers have been able to rapidly deploy virtual applications with integrated services by deploying VXLAN models to virtualized host servers.
Over the past few years, there has been an incredible increase in the popularity of VXLAN. The IP / UDP-based MAC overlays and support for up to 16 million virtual network segments have made VXLAN the premier choice for multi-user data center deployments.
In a status on the personal page, Ivan Pepelnjak, SDN Specialist, NFV, and author of several books on data center networking commented on the future of overlapping technology:
- FabricPath – A monopoly packaging technology Cisco is unlikely to have in the future due to its lack of support from multiple vendors.
- Even TRILL is an overlapping packaging standard of the IETF, but developed by only one vendor which is Avaya due to the disadvantages of Layer 2 packaging in general, while the advantages of VXLAN are obvious.
- ACI – Along with the much-promoted Fabricpath, there is still much work to be done to perfect this solution, along with the Cisco monopoly leading to restriction of dissemination across the entire data center system.
- VXLAN, with its advantages in both Unicast and Multicast communications, and the widespread availability of an open standard, will be the future of connecting and extending virtual server systems.
Clearly, VXLAN has shown superiority over the rest of the techniques, the fact that all device vendors and solutions that provide connectivity to the virtual server infrastructure have chosen VXLAN as a top prioritized packaging proxy. This is perfectly suited to the development trend in a flat, technologically changing world.
Vu Ba Dung – FPT ISRelated posts: