Facing challenges--Cisco technology innovation under the background of "cloud"

Cloud computing has brought unprecedented challenges to data center networks, which are mainly embodied in network scale, resource virtualization and sharing, and distributed data Center collaboration.

First look at the network scale. Cloud Computing Data Center Server size is usually thousands of units, or even tens of thousands of units, and with the development of the business, the scale of the network will continue to expand, with the attendant, cabling network management has become more complex. The traffic flow model is no longer the traditional north-south oriented, high convergence ratio, but the distributed cluster application represented by Hadoop and hive, and the flow model is mainly in the direction of East-west and low convergence ratio (near line speed). Traditional Data Center network architecture can not meet the needs, but it needs innovative system architecture, control plane, and a larger exchange capacity of the core switching equipment.

Secondly, virtualization technology is widely used in data center. Virtualization not only improves resource utilization, but also enables rapid and flexible resource deployment and scheduling based on changes in business requirements. Correspondingly, the dynamic change of the two-layer network environment required by the computing resources on demand scheduling (virtual machine migration) requires a large-scale two-layer networking environment, while the traditional data center two-layer technology has not met the requirements.

The characteristics of cloud computing require that business development has nothing to do with the physical location of computing resources, and requires that the computing resources between different data centers be integrated into a unified resource pool. How to ensure the efficiency of computing across the data center is also a new problem faced by cloud computing.

In view of these problems and challenges faced by the data center network in the cloud computing environment, Cisco has made technical innovations in several aspects to build a high performance, high throughput and highly scalable network platform for Cloud Computing data center.

Innovative Data Center network architecture--fabricpath

Fabricpath is an innovative technical feature that Cisco has implemented on the Nexus series of data center switches (NEXUS7000 has supported this feature and will be supported in the future Nexus5000) with the goal of implementing a two-tier environment, such as EMCP (equal-cost Multi-path routing, equivalent multipath routing, smooth expansion, fast convergence, prevent broadcast storm and loop, two-tier network configuration simple and easy to manage and so on.

From the protocol level, Fabricpath defines a new two-tier address space called the Switch ID, which is uniquely identified as a switch and provides a basis for routing addressing between nodes. A two-level TTL field is defined, similar to the TTL mechanism in the IP protocol, to prevent data frames from being forwarded infinitely in the Fabricpath domain.

Fabricpath introduces an improved is protocol as a control signaling, instead of relying on MAC address for addressing, and relies on switch IDs of switches to complete addressing, exchanging is signaling to build routing tables between nodes. The IS protocol also calculates the optimal path as the basis for data forwarding. The improved is protocol also implements 16 two-layer equivalent routing ecmp. This means that with a gigabit interface, the maximum bandwidth between any two points can be up to 2.56Tbps. High bandwidth also means that fabricpath can be applied to large scale data center networking environments.

In a virtualized environment, each VM (virtual machine) has a unique MAC address that causes the MAC address domain to become large, exceeding the MAC address capacity of traditional network devices. and Fabricpath uses "session based MAC address learning", only those destination address for the local server in the data frame of the source address will be put into the network device MAC Address table, the network device only save the session with the local MAC address table entries, This greatly reduces the Virtualization data Center Access device's MAC Address table.

As Cisco's important two-layer multipath technology innovation, Fabricpath represents the development direction of the next Generation data center network. At present, the two-layer multi-path standardization organization mainly has IETF,IETF Trill also uses is as routing protocol. Cisco plays a leading role in the development of trill standards, Fabricpath can be regarded as a "enhanced version of the Trill", is the trill of the basic functions of the ten "based on session MAC address learning", "vpc+ Multiple Topology" and other advanced features of the collection.

At present, Cisco's Fabricpath technology has commercial, Fabricpath domain can accommodate 12288 million Gigabit network card server, or more than 100,000 Gigabit NIC server.

Unified Exchange Matrix--Advantages of End-to-end network convergence

The new generation of Cisco Nexus 5500 up switch all ports can be designated as a Gigabit Ethernet, Gigabit Ethernet, FC (2/4/8g) or FCoE interface, this feature provides maximum flexibility for the data center. The Nexus5500 switch also supports the two or three-tier exchange, the IEEE 802.1QBH standard, and the hardware supports Fabricpath features.

NEXUS5500 's Matrix Extension Technology (FEX) ensures that users adopt a unified Exchange matrix (Nexus5500 internal Exchange matrix) to achieve the physical network card from the Nexus switch to the server, as well as the direct connection of the virtual machine, its design flexibility, simplification of wiring, and simple management, Can save a lot of equipment and operating expenses for the user.

In addition to the FEX Extender, which is equivalent to a remote extension module, Cisco has introduced Adapter-fex and VM-FEX technology to extend the unified Exchange matrix to the network connections of virtual machines. At the same time, the hardware exchange of virtual machine network traffic is realized, and the performance of virtual machine network I/O is enhanced. The virtual machine connection becomes a network interface on the unified Exchange matrix, which greatly reduces the complexity of the network management.

vn-tag--Virtual Machine Network access standard

Vn-tag is a standard proposed by Cisco for the Virtual Unit network, its core idea is to add a special tag--vn-tag in the Ethernet frame to distinguish the different virtual machine network interface, so as to identify the traffic of the specific virtual machine.

A new address type is defined in Vn-tag to identify the network interface (VIF) of a virtual machine. The vif of each virtual machine is unique. The Vn-tag of an Ethernet frame contains a pair of such new addresses dvif_id (destination vif) and svif_id (source vif) to indicate where the frame came from and where it went. When the data frame flows out of the virtual machine, a vn-tag tag is added. When multiple virtual machines share a physical link, the source address svif_id based on Vn-tag can differentiate different traffic and form corresponding virtual channel. This is similar to hosting multiple VLANs in a trunk link. The upper-top switch (for example, Nexus 5500) recognizes Vn-tag and generates the corresponding virtual interface Veth on the switch, which corresponds to the vif one by one of the virtual machine, as if the vif of the virtual machine is directly connected to the veth of the physical switch. All switching work is done on the top switch, even if the traffic is exchanged between different virtual machines within the same physical server, and is forwarded via the top switch. This approach, although increased network card I/O, but through the Vn-tag, the work of networking back to network equipment.

Cisco has presented the IEEE 802.1QBH draft based on Vn-tag as the basis for the next-generation data center Virtual access machine.

OTV Technology--two-layer interconnect technology between data centers

OTV is an innovative technology on the Cisco NEXUS 7000 switch, which is based on the idea of building an ip-based tunnel between the data centers, encapsulating two-tier Ethernet frames. The control plane adopts the MAC routing technology, learns the MAC address of other data center through the control signaling, and puts it in the local MAC address table. When you need to access a host in another datacenter, you can find the corresponding IP address of the data center of the other in the local MAC Address table, and then access it directly through the IP tunnel, as conveniently as it is for local visits.

OTV's advantage is that the operation is simple, do not need to change the network topology, and do not need to be like VPLS/MPLS to create a separate control plane, as long as the existing network to open the OTV function, you can achieve two-tier interconnection between multiple data centers, its implementation is quite transparent. At the same time, because the two-layer Ethernet frame is encapsulated in the IP packet, OTV can not only realize the ecmp of the two-layer exchange, but also avoid the risk of two-layer loop and broadcast storm.

The OTV realizes the two-tier interconnection between data centers, and facilitates the implementation of collaborative computing and virtual machine migration across data centers.

In short, Cisco believes that cloud computing will be the next phase of computing and the evolution of the Internet. Cisco's cloud strategy is to be user-centric, the network as the operating platform for the cloud, to ensure that users through the platform in collaboration, innovation and security to carry out business, and further accelerate the development of cloud computing business. Around this strategy, Cisco through a series of innovative technology to ensure that the data center network to adapt to the development of cloud computing business requirements.