——Huawei Cloud Fabric Solution

　　As cloud computing services develop rapidly, data center network architecture must evolve and adapt in turn. Based on its experience, Huawei promotes the cloud fabric solution to adapt to the cloud era.

　　Cloud Data Center Network Development Trends
 

　　Demand for high-performance computing, data mining, and data storage is increasing, and data centers are expanding to meet that demand. Supporting more cloud computing applications, more servers, and more storage devices is a tremendous challenge for data center networks. Customers need a cloud network that is highly scalable and easy to use, and supports on-demand resource allocation.

　　Three trends will guide the development of next-generation cloud computing data center networks:

　　Trend 1: large-capacity networks, high-performance core switches

　　IDC predicts that 10GE servers will be mainstream in data centers beginning in 2012 and that 40GE/100GE servers will become widely used within the next five years. A Cisco report estimates that global cloud computing traffic volume will increase twelve-fold between 2010 and 2015, with a compounded annual growth rate of 66%.

　　To handle this explosive growth in traffic volume, data center networks must have extremely high capacities, and core switches must provide super high switching capacities and port densities. Most of the data center core switches in use today were designed 3 to 5 years ago for GE servers. The backplane bandwidth, switching capacities, port densities, and power consumption of these core switches do not meet the requirements of the 10GE servers that are replacing GE servers. Data center networks need a next-generation hardware platform and advanced technologies to support the 10-year evolution path of servers from 10GE to 40GE and finally to 100GE, so that the network architecture does not need to be changed frequently over the long term.

　　Trend 2: virtualized networks, flexible resource scheduling

　　Cloud-computing data center networks require virtualization technologies for the following reasons:

　　1. Virtualization technologies are needed to adjust network boundaries. For example, 1:N virtualization allows network resources to be allocated on demand so that physical devices can be used more efficiently. N:1 virtualization simplifies the logical topology of a network, and this reduces network operation and maintenance costs.

　　2. Networks must be virtualized to support server virtualization. According to a survey by Gartner, nearly 50% of servers today work on virtualized platforms. The flexible deployment and migration of cloud applications has become a mainstream requirement in the cloud computing era. Data center networks must be virtualized to support online migration of virtual machines. Newly developed multi-tenant applications require a large network-wide resource pool to provide services for many tenants. A data center network must be virtualized so it can isolate and control services for different tenants and allow flexible deployment of virtual machines for tenants.

　　Trend 3: converged networks, unified switching platforms

　　A data center network allows resources to be exchanged between various network devices, such as between multiple servers or between multiple storage devices. Historically, a data center has consisted of multiple independent networks, including an IP data network and an FC storage network. This network architecture becomes increasingly problematic as a data center expands. For example, each server needs multiple dedicated network adapters, which increases both power consumption and hardware costs. Equipment rooms also need several cabling systems, which complicates network management and maintenance.

　　Industry-standard Fiber Channel over Ethernet (FCoE) addresses these issues. Ethernet not only provides high bandwidth but also ensures low latency and low packet loss to ensure lossless transmissions over the storage network. FCoE switches are replacing FC switches in next-generation data centers, allowing all FC and IP data services to be transmitted on an Ethernet network. This converged network reduces power consumption and lowers network construction, management, and maintenance costs.

　　The core requirements for a cloud data center network are scalability, virtualization, and network convergence.

　　Huawei Cloud Fabric Solution

　　Huawei released CloudEngine (CE), a series of next-generation data center cloud switches, at the Interop Expo held in the United States in May 2012. The CloudEngine series includes CE12800 flagship core switches, which feature the highest performance in the world[JB2] , as well as CE6800/5800 high-performance box switches.

　　Huawei also launched the Cloud Fabric solution for next-generation cloud computing data centers. This solution combines CloudEngine series switches with Huawei transmission, routing, security, and network management products to help customers build a stable network architecture supporting development for the next 10 years.

　　Huawei Cloud Fabric Solution Key 　　　　　　　　 Features

　　Huawei cloud fabric solution helps enterprises and carriers build a 360 Tbit/s non-blocking switching platform that provides access to tens of thousands of 10GE/GE servers. This platform has three major advantages: scalable fabric, virtualized fabric, and converged fabric.

　　Scalable Fabric
 

　　Scalable network to support development for the next 10 years: CloudEngine switches can be used to build a 360 Tbit/s non-blocking switching network that supports the evolution of four generations of servers (from GE/10GE to 40GE/100GE). With this non-blocking switching platform, customers do not need to upgrade their networks to meet requirements for 40GE access or even 100GE access as frequently as before.

　　Highest performance core switch in the world: CE12800 provides 2 Tbit/s of bandwidth (unidirectional) per slot and 48 Tbps of switching capacity. It supports various high-speed line cards including 8*100GE, 24*40GE, and 96*10GE line cards. One switch can provide a maximum of 96*100GE and 288*40GE full-line-speed ports or 1152*10GE full-line-speed ports. The large capacity and high port density make the CE12800 an industry leader.

　　40GE uplink ports on all TOR switches in the whole series: CE6800 and CE5800 provide super high densities of 10GE and GE access ports. All of the TOR switches across the whole CE6800/5800 series provide 40GE uplink ports for the highest forwarding performance in the industry.

　　Non-blocking CLOS architecture: CE12800 core switches use a non-blocking CLOS architecture and dynamic distributed buffering technology (100 ms per port). These technologies can handle incast traffic resulting from cloud computing applications such as search service and parallel computing, ensuring zero packet loss during service transmission.

　　Highly reliable redundancy design: The CloudEngine series has a high rate of reliability to ensure nonstop service transmission. All key components of the CE12800, including the main control units, switch fabric units, power modules, and fans, work in redundancy mode. CE6800/5800 switches also have hot swappable power modules and fans working in redundancy mode.

　　Strict front-to-rear ventilation channel design: The whole CloudEngine series uses a front-to-rear ventilation channel design, which is the most appropriate for data center equipment rooms. This design separates cold air flows from hot air flows to improve heat dissipation efficiency.

　　Virtualized Fabric
 

　　Comprehensive network virtualization capabilities: 1. The CloudEngine series uses Virtual System (VS) and Cluster Switch System/intelligent Stack (CSS/iStack) technologies to flexibly adjust network boundaries, meeting requirements for network virtualization. 2. The CloudEngine series uses Transparent Interconnect of Lots of Links (TRILL) and nCenter technologies to support service virtualization and server virtualization.

　　VS to achieve on-demand resource sharing: The CE12800 uses VS technology to provide an industry-leading virtualization capability of 1:8. One switch can be virtualized into a maximum of eight logical switches, allowing one core switch to manage services of different areas (such as production area, office area, and DMZ) or multiple tenants. VS technology improves device usage efficiency and reduces network construction costs while maintaining a high level of security.

　　CSS/iStack to simplify network management: The whole CloudEngine series supports CSS/iStack, which can virtualize multiple physical switches into one logical switch to simplify network management and improve network reliability. The CE12800 series has an industry-leading clustering capability, wherein four CE12800 core switches can form a cluster. The CE6800/5800 series supports a large stack of up to 16 switches.

　　Innovative CSS+VS synergy technology: The CE12800 combines CSS and VS technologies to turn a network into a resource pool so that network resources can be allocated on demand. This on-demand resource allocation is perfect for the cloud computing service model.

　　Large-scale routing bridge to support flexible service deployment: All CloudEngine series switches support TRILL, a standard IETF protocol. The TRILL protocol helps build a large Layer 2 network with more than 500 nodes, which allows flexible service deployment and VM migration in a wide range. 10GE/GE servers can also be used on a TRILL network.

　　nCenter to allow auto migration of VMs: nCenter is an automated management system for virtualized data centers. It automatically delivers network configurations to devices when VMs migrate. Additionally, nCenter provides high-speed RADIUS interfaces to increase the VM migration speed to 10-20 times the industry average.

　　Converged Fabric
 

　　Converged network to reduce TCO: The CloudEngine series supports FCoE, which enables a single network to provide storage, data, and computing services, reducing network construction and maintenance costs.

　　Centralized FC gateway deployment to facilitate O&M: In centralized FC gateway deployment, the Ethernet network is compatible with the FC SAN network, but Ethernet access devices do not participate in computing of the FC protocol. Centralized FC gateway deployment makes network O&M simpler.

　　Low latency and zero packet loss: The CloudEngine series supports Data Center Bridging (DCB) to ensure low latency and zero packet loss for the FC storage service and high-speed computing service.

　　Summary
 

　　Huawei has 20 years of experiences in the IP field and the largest data center product portfolio in the industry. Huawei offers end-to-end data center solutions covering network infrastructure, disaster recovery, security, and network management. Huawei data center products and solutions have been widely used in large enterprises, vertical industries, Internet corporations, and carrier networks.

　　As one of the world's leading network solution providers, Huawei has an excellent long-term plan for data center network development and a firm determination to invest in the data center field for the long run. Most importantly, Huawei has world-leading research capabilities and experts, rich experience in the pre-research of data center standards, and chip development capabilities.

　　The CloudEngine series switches and Cloud Fabric network solution are products of Huawei's rich experience working with IP networks during the last decade. They will help customers build a next-generation data center cloud network to support the sustainable development of cloud services in the future.