The evolution of large data center networks: Meeting the T-Class age

The achievements of modern science and technology make human organs in space has been greatly extended, if the telephone is the extension of human sound, television is the extension of human vision, then the large data center network is the human Brain unlimited extension, large data center switch is the brain nervous system as a high-speed road. The normal nervous system is one of the hallmarks of a person's health, and similarly, large bandwidth, low latency, and high reliability of the network system is the basic guarantee for human to enter the cloud era.

Data center is the inevitable outcome of internet development to a certain stage. Relevant data show that with the advent of the cloud computing era, the network traffic model has changed greatly, 16% of traffic is concentrated in the operator network, 14% in the corporate network, and 70% of the traffic will be all flow to the internal data center. The exponential growth in data center traffic and bandwidth is far beyond the imagination of traditional networks. Large bandwidth, high scalability has become the primary demand of data center customers, the construction of data center has entered a new era of T-class.

On the development trend of data center, it may be useful to review the development process of Ethernet. Ethernet after nearly 30 years of development, bandwidth from 10M onwards, experienced 100M, 1000M, 10G, 40G, 100G, the development phase of the current 10GE Ethernet has been batch applications, 40GE and 100GE Ethernet began to gradually apply. The development of Ethernet is actually witnessing the rapid growth trend of data center traffic. From the early X86 server to virtualization, the gradual increase of virtual machines and the increasing frequency of data interaction between virtual machines make server's demand for bandwidth more and more strong; When the data center starts to enter the cloud computing age, the data center server, Storage and all IT infrastructure resources in the network tend to be virtualized, which means that the entire data center needs strong network bandwidth to meet this trend; Now the data center is in the middle of a big data era as it approaches the evolution of cloud computing data centers, If cloud computing makes the resource in the data center highly virtualized, thus dramatically increasing the bandwidth requirements of the data center, the advent of the large data age has made the data center's data source vastly wider, data-device access diversified, data capacity vastly larger, processing needs to be faster and more efficient, All of these undoubtedly put forward a higher level of demand for the network of data center.

Data Center Network Architecture Evolution Trilogy

We all know the importance of data Center network architecture in the process of data center development, and how to measure the network architecture of data center needs to be judged by the merits of the core switch of data center. Below, we analyze the development trend of data center network by reviewing the development of core switch of data center.

The first stage, high-density fe/ge access convergence, a small number of 10GE uplink. This phase of the core switch to the central Exchange and LSW mainly composed of central exchange mainly Ethernet switching chip, card board to Fe/ge and a small number of 10GE LSW as the interface chip, the line card provides 48GE wire speed forwarding, business processing basic to two and three-tier, mainly, QoS processing is mainly based on simple priority queue scheduling. Typical products are cisco4500/6500, h3c6500/7500, etc.

The range of applications for this type of product satisfies the Fe/ge access and a small number of 10GE uplink scenarios, with a small data center network consisting of a small number of servers in the context of a GE server-driven scenario in the data center domain.

The second stage, high-density ge/10ge access convergence, a small number of 10ge/40ge uplink. At this stage, there are many forms of central exchange, such as simple Exchange network with shared cache, some centralized arbitration exchange network, and 6.25G Clos Exchange architecture. The card single board takes GE and 10GE LSW as the interface chip, a few 40GE upstream interface mainly. The bandwidth of each line card slot is less than 480G, the business processing basically is two layer and three layer mainly, has the simple Hqos dispatch, the cache scope is very wide, relies on LSW chip, some switches are 2ms/ports, some 10ms/ports. Typical products are Cisco N7000, h3c10500/12500, etc.

The range of applications for these products meets the GE Access and 10GE uplink scenarios, which meet the central and Small Data Center network of GE servers in the case of a GE server-driven scenario in the data center domain.

The third stage, high-density 10ge/40ge access convergence, a small number of 100GE uplink. This phase of the core switch to Clos dynamic routing switching, the card single board with complex business processing capacity of the PP chip, to provide high-density 10ge/40ge wire speed forwarding and business processing, a small number of 100GE upstream interface mainly. Back-plate link to 10G start, the future can evolve to 25Gbps, each line card slot bandwidth is 1t~4t, the current maximum capacity 48~96*10ge wire speed forwarding, or 24*40ge wire speed forwarding, with perfect QoS processing capacity, can support a larger cache, can reach 100ms/port. Typical products are Huawei CE12800 Data center switches.

The application scope of such products to meet the high density 10ge/40ge access scene, in accordance with the evolution of the server, starting from 2012 will be mainly 10GE server, in 2015 gradually 40GE server, so this phase of the data center core switch must be able to provide high-density 10ge/ 40GE interface to meet the large and Medium data center network composed of 10GE/40GE servers.

T-Class era: How to grasp the lifeline of data Center network development

When we see the huge data centers owned by users such as Google, Facebook, Amazon, Baidu, Tencent, and Alibaba, we realize the impact of trends such as cloud computing, social networking, big data, and mobility on data center development. It is with these transformative waves that the time span between the bandwidth of each generation of Ethernet is drastically shortened, and a few years ago our server 10GE ports were rare, and now the latest is the standardized 100GE port. As a result, a new generation of core network switches for cloud computing and large data centers must have a strong bandwidth and capacity capability to support high-density 40GE and 100GE interfaces. "The Enterprise park market is relatively stable in view of regional differences, and the data center continues to respond to a surge in shipments of 10G and 40G switch ports," said Rohitmehra, vice president of IDC network Infrastructure. ”

In fact, we can verify the future direction of the data center's network development through a product that best represents the future trend. Huawei's CE12800 Data center switch for cloud computing provides 2t-4t/slot bandwidth scalability, and uses a rigorous front and rear heat ducts, 1000W of slot cooling capacity, and provides up to 192 100GE, 384 40GE and 1536 10GE full speed ports, maximum support 128Tbps backplane Exchange capacity, with perfect QoS processing and at least 100ms large cache capacity per port. Huawei CE12800 Data Center core switches in the industry for the first time to achieve the data center 1T capacity card of large-scale commercial, commercial process far more than other mainstream manufacturers, leading the Internet network equipment capacity farewell 100G, 400G era, into the T-class new era.

There is no doubt that "born in sorrow, euthanasia" is the most reliable survival rule in the era of innovation and change, data center as the core support and strength of enterprise business development, must have the ability to adapt to the future development. In the era of cloud computing and large data, T-class or data center building standard, with a solid, reliable, for the future and the design of the core data center switch can accurately grasp the data center network Development Lifeline, and then protect enterprises in the future competition in an invincible position.