Six things to figure out before cloud service.

What are the real concerns of data center users before the cloud transition? What kind of cloud technology should the data center look for in the face of different service requirements? What are the key technologies that determine the efficiency, energy consumption, and management of cloud computing data centers? Through cloud practice in China's local data centers, Let's look for answers to these questions together.

"The past year has been an exciting year, and the development and value of cloud computing has been like the role of electricity when refrigerators and televisions come into play." "In the face of the rapid changes in China's cloud services at the China Cloud Computing Conference this year, Jian, chief technology officer of Alibaba Group, has made this feeling."

2013, China's cloud computing industry has entered the growth period, based on cloud computing from the point to face of innovative applications permeate all fields, finance, transportation, electricity, manufacturing and other industries have touched the "cloud" to seek breakthroughs for the Internet and intelligent city applications launched cloud Computing data Center building in full swing. From the Aliyun to Amazon, Tencent Cloud mature, to cloud storage, cloud manufacturing, cloud management services, and then to the government cloud, education cloud, financial cloud, urban management cloud, health care cloud, location services cloud and so on a series of cloud applications rich, all show the Chinese cloud service market huge potential and broad prospects. According to IDC, a market research firm, China's cloud services market will reach $1.1 billion trillion in 2014 and will grow at a compound growth rate of nearly 45% over the next 4 years.

Nowadays, more and more enterprises see the opportunities and changes brought by cloud service, but there are still a lot of enterprises puzzled about how to set up the carrier of cloud service-data center. Many CIOs say they have a clear expectation of the service capabilities of the data center, but they don't know how to make sure that the cloud-centric data center has the ability to serve the applications it is hosting and really play the role of the cloud.

What are the real concerns of data center users before the cloud transition? What kind of cloud technology should the data center look for in the face of different service requirements? What are the key technologies that determine the efficiency, energy consumption, and management of cloud computing data centers? We may as well come together from the experience of the pioneer of cloud practice in China data Center, Look for answers to these questions.

Ignore this, cloud computing will be a piece of paper

In 2009, Intel's IT department is trying to transform its global data center into a private cloud architecture in order to "improve employee productivity, help business units develop, provide flexible delivery patterns, and support new businesses." By deploying a private cloud, Intel saved $9 million trillion in spending over the past 5 years; Reduce the time required to apply for infrastructure services from 14 days to 45 minutes, improve performance by 5 times times while reducing the 40% design servers, and allow 80% of new services to be deployed in the cloud and provide self-service services. So far, Intel's Shanghai Purple Bamboo Data Center has been nearly 75% virtualized, and the use of blades for EDA design is up to 80%.

In just a few years, the complex IT systems that support nearly 20,000 employees around the world can be used in the process of building the cloud, what is Intel doing to build the clouds by translating them into efficient private clouds without impacting the business, and translating the value of the cloud into the value of dramatically reduced costs and increased efficiency?

"The goal of cloud computing is to achieve a double rise in business efficiency and user experience, ignoring these two points, cloud computing will be a piece of paper." At the same time, it is an important part of cloud computing deployment to implement cloud management effectively and build a computing environment with different security levels according to the different nature of the industry. "When it comes to Intel's experience with building a private cloud, these are the basic principles that Intel's IT department references."

Based on its long-standing experience in data center construction and computational load, Intel it has developed a "balancing act"-balancing the balance between improving computational load and reducing resource consumption, achieving management needs and reducing capital investment, sound development and positive innovation to enhance competitiveness. Intel IT department has also developed three major strategies for this purpose: first, the continuous optimization of the data center, covering the server, storage, network, infrastructure and other levels, so that it and business development to match, and the second is to improve data center utilization, that is, through business optimization and technical means to maximize the use of existing infrastructure, reduce resource input; The third is to implement the regional planning strategy of vertical integration of data center on the basis of considering local demand.

When planning data centers, companies are often used to focusing on the technology architecture of the cloud, such as how to implement virtualization, which is a myth. The process of transforming an enterprise data center into a cloud is not only a change in the IT infrastructure, but a radical change in the way IT services are made.

The traditional data center is more to support enterprise business, while the core value of the cloud is much more than this, enterprises use the driving force of cloud is to enhance business efficiency and user experience, support globalization, mobile change. So before planning the datacenter, it is precisely the basic principles, implementation rules and strategies considered by Intel IT departments to consider the transformation of the architecture of the data center in such a way that it is possible to "not be confused" and "not biased" in the direction of investment construction and technology selection. The "balance rule" and the three implementation strategies, based on Intel's experience, are now highly valuable for all enterprises planning or building private cloud data centers.

The key to transition: a unified network architecture

The transition to a data center cloud is not just about how to make computing or storage resources realize pooling. Because virtualization allows computing and storage resources to be pooled, allocating and delivering these resources is done over the network. In such a data center, the function of the network has undergone great changes, it must have the bus-like efficiency in order to successfully deliver the cloud services. Therefore, in the process of transforming traditional data center into cloud, the transformation of network function is the key of data center transformation.

Three years ago, the trend of data and storage network convergence began to change data centers. FCoE as the key technology to achieve this convergence is highly respected by it vendors, Intel also implemented the LINUXS kernel of open source FCoE, and further promote the technology in the data center wide application. With the price of gigabit Ethernet ports falling, the integrated network has begun to serve as a cloud data center in the cloud practice of many domestic users.

Data center of Shanghai Jiaotong University in the process of cloud transformation, the data and storage network Fusion scheme was enabled, and they realized the smooth transition of traditional FC storage network to FCOE storage network with the help of new FCoE technology. The experience is that Cisco Nexus 5500 in the data server area enables FCoE functionality, some interfaces are configured with 8G FC SFP modules, and the SAN storage system chooses the NetApp FAS series Storage System that supports FCOE. NETAPP FAs and the server's gigabit NIC are connected to the Nexus Switch's gigabit interface, so that on the same 10,000 gigabit ports, the FCoE, iSCSI, and NAS protocols can be supported, and the original FC-configured storage devices and servers are connected to the Cisco Nexus 5500 FC interface. and FCoE network integration. NETAPP FAs FC interfaces access existing FC switches to provide storage space for existing FC Sans. In view of the strong exclusivity of the FC switches of each manufacturer, Shanghai Jiaotong University did not interconnect the existing FC switches with Nexus 5500. In their view, over time, FC SAN devices will naturally be eliminated and the data center will transition to pure FCoE storage. It is expected that more than half of the servers in its data center will be configured with a gigabit fusion port in the next two years.

Cloud data centers need a unified network architecture more than ever, mainly because when virtualization transforms the data center into a cloud environment, the heterogeneity of the transport network and the storage network creates more complex traffic. More and more complex traffic needs to be managed, which is bound to turn the cloud into a inefficient system. Especially with the increase of traffic scale, it is very difficult for data center to guarantee usability without reducing performance. At the same time, heterogeneous networks can be more costly when they need to be expanded. Only by adopting a unified architecture, can we achieve the goal of simplifying the network architecture and simplifying the operational dimension to enhance efficiency.

The transformation of the Software definition data center

The cloud of data center into the design of the datacenter architecture poses some new problems: if the likelihood of the future of such a data center is almost zero, what kind of architecture and technology will make it sustainable? As data centers transmit, collect, summarize, and analyze the volume of data that is growing, the network, How should the new challenges of computing and storage be addressed?

The technological evolution of server virtualization, storage virtualization, and network virtualization does not happen by chance. Virtualization makes IT infrastructure flexible and scalable, and it is proving that the agility and resilience of business can only be based on a flexible IT infrastructure in the face of a complex and changing business environment and the development of the Internet. The development of data centers will be service-oriented, and must be easier to manage and operate.

So the development goal of the cloud data center is to become such a data center-software-defined datacenter, a data center that is more flexible, smarter, and less effective than the physical hardware. The transformation of data center must be around the network, transformation of automated processes for server and storage: The network evolves from a storage area network share to a "storage-as-service" cloud storage that will evolve into a SDN-based service-aware network, and will be evolved to be resilient in terms of load requirements and must be secure, Credible。

Of course, the direction of this transformation has also led to a new idea, that is the future data center will evolve into computing, storage, network resources seamless integration? At present, Intel is exploring this. For software-defined data centers, Intel has proposed a new architecture-SDI (software-defined infrastructure), SDI's goal is to allow future open Architecture data Center IT resources from static to dynamic, and its management from manual to automated. Its core idea is to let the computing, storage, network through a unified computing architecture to achieve the effectiveness of the release, no longer need to use a variety of optimization techniques to make up for different structures caused by internal friction. Based on this unified architecture, cloud data centers can quickly and efficiently implement resource automation rationing and management capabilities.

In large cloud computing data centers, the transformation of computing, storage, and network physical integration has begun. The best example is Intel's new generation of Cabinet Architecture (RSA), which is valued by builders of large cloud computing data centers such as Baidu, Alibaba, Tencent and China Telecom. This new rack structure, instead of using Fibre Channel, PCI ports, or iSCSI interfaces to connect storage units within each rack, you can significantly reduce the use of cables by using Intel high-speed silicon photon technology to connect storage units within each rack, with more than 100,000 server volume data centers. Reduce costs. More crucially, it allows for a significant increase in deployment density relative to the traditional server architecture, resolving the high density deployment challenges experienced by large cloud computing data centers.

How to Look at SDN

According to IDC forecasts, over the next 5 years, more than 2.5 billion users will be connected to the Internet, the number of devices used will be more than 10 billion. Compared with the current situation, it is expected that by 2015, this development will require 8 times times the storage capacity, 16 times times the network capacity and 20-fold computing capacity. Has today's data center builders taken into account what kind of disaster this pressure will bring to the data center? After the network integration, how should the network evolve to cope with such traffic, the explosion of data growth?

Bongjunying, a senior expert at Ali Group Network data Center, believes that a cloud data center network should have at least the following characteristics: streamlined operations, rapid deployment efficiency, standardization, and Green. To this end, Aliyun has built a standardized, modular cloud Datacenter Network architecture.

It is noteworthy that today Aliyun has seen the network as Ali to the well-off "road." Why does Aliyun have such a knowledge of the Internet? Because Aliyun has seen Sdn. After the network fusion, rely on SDN, Aliyun will be able to according to the needs of the business free definition network ability, let network efficiency completely release, for its business to provide efficient support. Bongjunying that Aliyun's data centers are trying to leverage SDN to support the business. The value of SDN has been favored by many operators of cloud data centers, such as the era of interconnection in the construction of a SDN based operating model.

Before building a SDN network, there are many technologies and platforms that data center users need to focus on. In fact, SDN will no longer be a problem in the field of network communications. From the point of view of technology, the biggest change brought by SDN is that the network control function is separated from the network equipment of forwarding packet, and the control function is implanted in the software of the standard computing platform, so the computing platform becomes a key point to decide the SDN development.

It can be seen that the current network vendors are mainly based on the private computing framework to promote SDN, the user's independent development is relatively difficult. To support the development of SDN, Intel has also introduced the Open Network Platform Switch design architecture, open Network Platform Server design architecture, and also launched the Data Tier Development Kit (Intel Data Plane Development Kit,intel DPDK), To drive the entire biosphere to accelerate the development of SDN hardware and software products. It is reported that only with the DPDK software library, packet processing performance can be increased by up to 10 times times, so that the server while processing the packet also through the Intel processor to perform other workloads, ultimately reduce hardware costs, simplify the application development environment. With the help of such development kits as DPDK, the time for data center users to develop new business based on SDN can be dramatically shortened. So, at present DPDK also received a lot of SDN developer's hot hold.

The storage required by the cloud

Under the demand of cloud computing and large data, how to store, Dispose and use of massive data has become a hot topic in all industries. In a highly virtualized environment, the efficiency of computing can be improved, but still with traditional, rigid storage architecture processing data, it is difficult to avoid cost and time double loss situation.

Three new features of the industry-defined storage of software-automated tiering, automated data management, and massive scalability to make data storage more responsive to cloud storage and large data requirements.

It is worth noting that many problems faced by traditional storage systems can be solved in the improvement of computing platform. With the advent of the Intel Xeon platform as the next Generation storage solution optimization technology, the scalable intelligent storage solution has resolved many storage challenges for large data applications. For example, by automatically streamlining configuration, data de-duplication, and so on, to allocate space when data is written to disk, to solve the waste problem of disk space by means of indexing redundancy and duplicate data, to meet the changing requirements of the business to the cloud.

For large data applications, Intel also introduces many intelligent storage technologies and tools to help businesses reduce the amount of data they need to store and improve the way they use data. For example, the streaming of cold and hot data can improve the cost of data storage, and improve the data analysis ability greatly for Hadoop optimization.

In addition, Intel is promoting the innovation of solid-state hard disk technology and the application of such storage devices in the cloud data Center for the problems of traditional hard disk Gao Shiyan and high error rate. As the price of SSD equipment drops, a series of solid-state drives with high bandwidth, low latency and high IOPS will improve the storage problem of cloud.

Virtualization and security issues

In many cloud data center practices, high levels of server virtualization are rare. In fact, in addition to the functionality provided by virtualization software, if you are good at using the hardware platform virtualization optimization capabilities, the effect of virtualization may be far more than you expect. Currently, Intel provides Vt-x technology to support 32-bit, 64-bit virtualized environments based on multi-core Intel Xeon Processor-based servers. By extending the core platform architecture, Intel Virtualization technology helps users bridge the gap and eliminate barriers in the current virtualization solution. One of the improvements that this architectural extension can achieve is: The new higher privilege ring for virtualization software (CPU access privilege level), with a higher privileged ring, enables client operating systems and applications to operate in the applicable ring while ensuring that virtualization software has privileged control over platform resources. It eliminates many potential system conflicts and simplifies software requirements.

Security is a sharp sword hanging over the head of cloud services, because the cloud extends the management boundaries of the enterprise beyond the data center, leaving the traditional protection of data center data and platform unprotected. How do you make the cloud a trusted system? The answer is to build a trusted pool of server resources.

In fact, an enterprise can use a virtualized server's trusted computing pool to build trust in a dynamic environment that can be trusted by external entities based on known trusted signatures. A trusted computing pool can provide a degree of security to a virtual machine before it is transferred to the platform and support virtual machine migration between trusted computing pools. The protection of such workloads can be enhanced by putting those sensitive workloads into the trusted computing pool.

Building a trusted computing pool is an important part of the security in private cloud practice, and it is also a necessary measure to ensure security in data center operation.

The misunderstanding of energy consumption management created by "theory"

The most important problem in the development of the cloud of data center is the impact of energy management on operating cost. Perhaps some may ask, is high level of virtualization not a magic weapon to solve energy efficiency problems? In theory, high levels of virtualization can actually help data centers reduce physical devices, reducing energy consumption. But in the data center, the theory is hard to achieve. Because simply using fewer physical hardware devices to encapsulate virtual machines is not enough to achieve the goal of lowering electricity tariffs. If the integration measures are not implemented properly, the energy consumption of the data center will even increase.

In the cloud computing datacenter, reducing energy use and energy costs is not proportional to reducing overall energy load. The reason for this is that when the physical server is consolidated onto a virtual platform, the process of transferring the workload leads to a drop in I/O capability, which can lead to performance degradation. When performance drops, the time required for the system to successfully complete the task increases and the energy consumption increases. At home, the company's cloud computing practices, such as the 600410, shares, have encountered similar problems. This problem can only be alleviated if the calculation performance increases and the I/O throughput rate increases dramatically.

The Intel Shanghai Bamboo Data Center accommodates more than 330 servers in 340 square meters of space, but the pue value of the datacenter can be reduced to 1.3.

Intel has successfully promoted energy efficiency by 2:1, the next generation Xeon processor itself in the I/O and energy consumption of a significant improvement, so that the data center in increasing the proportion of virtualization integration, on the basis of improved energy consumption. Huasheng Tiancheng is also in this way to solve its cloud computing data center energy efficiency problem. Second, intel on the data center of Thermal energy-saving technology, a lot of attempts have been made: they isolate the hot air duct from the cold wind, adjust the cable layout in the data center, and improve the overall temperature from 18 degrees Celsius to 22 degrees while ensuring the normal operation of the data center; bold to enable natural wind cooling- When the outdoor temperature is below 18 degrees, the natural wind is used to cool it.

While it is not unrealistic to gain energy efficiency gains as the computing platform evolves, while large data centers cannot rely solely on this idea to control energy efficiency, Intel believes that only policy-based energy management solutions can deliver rapid rewards for data centers of this size. It is reported that Intel it is in the use of Intel's Intelligent Power Node manager and data Center Manager to achieve a single server and server node power consumption real-time monitoring and control. With I/o-intensive workloads running, server energy consumption can still be reduced by 20%.

In the process of data center cloud, enterprises will gradually feel the computing, network, storage technology changes will bring about the effectiveness of the cloud. In the foreseeable future, it is easier for companies to control the cloud energy in the data center when they see the changes that are bound to occur in the data center evolution earlier.