New network technology in cloud computing

Since 2006, when Google introduced the concept of cloud computing, as a new technology and business model, the concept of divergent views. Various agencies and experts in related fields have defined cloud computing from different perspectives, such as some from the application scenario, and some from the resource angle.

Among them, the September 2009 U.S. National Institute of Standards and Technology (NIST) proposed the definition of cloud computing, very representative, has been widely accepted by the industry: "Cloud computing is a convenient access to a shared pool of resources, on-demand computing resources (these resources include network, server, storage, application, Services, etc.). Resources in a shared resource pool should be quickly deployed and published with fewer administrative costs and simple business interaction processes. "In addition, NIST is defining the deployment scenarios, resource types, and technical features of cloud computing as an extension of this concept."

In NIST's definition, "network" appears directly two times, but the two times "network" in the cloud computing role significantly different. In addition, the third role of the network in cloud computing is hidden by NIST in the "resource pool" it mentions. In the conceptual extension of cloud computing, "Broadband access Network" is one of the 5 basic characteristics of cloud computing.

The first time a "network" appears in the definition is responsible for connecting the cloud's service providers and users to make the cloud services available everywhere. The "Broadband access Network" in the cloud computing concept extension is also equivalent.

From the scope of application, the network can be public Internet (Internet), dedicated IP network or virtual private network (VPN private receptacle). and access technology, can be wired (such as xDSL, digital loop route), can also be wireless (Wireless LAN or 3G network). The idea of mobile cloud computing is also said from this perspective.

Viewed from a user terminal, it can be a generic soft terminal (such as a browser) or a dedicated soft terminal (such as some cloud security applications that require users to download specific client software), or even hardware-related (such as Apple's recently launched icloud, can only provide synchronization services for user data between Apple series products.

The new requirements of cloud computing for the "web" here, with the "Future Network", "Next generation of Internet" and "Next Generation Network" and so on, has been the pursuit of the goal is basically consistent, that is, the need to provide higher quality assurance network services, more secure and reliable information transmission, more energy-efficient green services and ubiquitous services.

The second appearance in the NIST cloud definition of "network", itself and computing, storage, and so tied into a resource, a cheap and fast to provide users with "network" resources. From this definition, IP VPN is a typical cloud computing service: The use of MPLS/IPSEC/L2TP and other tunneling technology, through the IP network (IP private network or the Internet) for enterprises or individual users, to provide quality assurance, security and credibility of the network resource rental services.

Cloud-oriented VPN business is the development of traditional IP VPN business, to those who want to outsource software, platform, computing or storage from the data center to provide flexible network resource rental services. The typical feature of these VPN users is that they do not want to use the public Internet to access the cloud Computing data center, but they want more and more accurate sharing of cloud computing datacenter Resources, such as virtual machines, physical servers, and the ability to control end-to-end resource usage. Cloud-oriented VPN extends the traditional VPN endpoints from the network edge of the WAN to the servers in the data center and even the virtual servers, making the data center's various cloud computing services a new VPN attribute.

Third, the "resource pool" involved in this definition also implies a third understanding of the "network", which connects various resources to form a "network" of shared resource pools, primarily the network within the Internet Data Center (Internet datacenter, IDC). At present, there are three kinds of network in IDC: Connecting the server computer networks, the bottom main use is the InfiniBand technology, the storage area network that connects the storage device, the bottom main use is the fiber channel technology, the IDC communicates with the external communication network, the bottom main use Ethernet technology.

As the cloud computing, storage and network are unified as a resource to treat, IDC's "triple network Convergence" must also be achieved. Cloud computing data centers pursue a unified network architecture, unified technology and unified management to simplify systems and reduce the security risks and vulnerabilities of complexity. From the current development, IDC's underlying network technology will be unified to the new Ethernet technology. Ethernet originates and succeeds in LAN, after the intrusion into the WAN, metropolitan area Network and WLAN, the last domain has been entered: IDC (the first kilometer), and technology has been developed from the latest to the Fusion-type enhanced Ethernet, which can support the storage network and the server network.

The "triple net convergence" of the data center has just started, so how low-cost, large capacity and non-blocking, the cloud Computing data Center to connect various resources to form a pool of resources, the interests of the industry and controversy is very large, highlighted in two aspects: first, the data center of large-scale network, and the second is the virtual Access data center.

In large-scale networking, the IETF advocates trill (transparent interconnection of the Lots of the Links (Trill)) protocol, while IEEE advocates an extension of STP (spanning tree Kyoto). Trill mainly for the shortcomings of Ethernet STB protocol, that is, STB can not take advantage of the private access between the switch, so the route is not always the shortest and fastest, and the convergence time is relatively long, directly lead to poor scalability, as well as the link fault is more sensitive. IEEE's Shortest path bridging technology is an extension of the multipath STP, and also uses link state technology to allow the switch to learn the shortest path and dynamically adjust the topology. Both technical standards have strong support from industry giants, while other giants either stand or support both agreements. It is worth mentioning that many enterprises on these two outstanding standards, unexpectedly have done their own private expansion.

In terms of virtual access, there are already two solutions, one is to add a special tag (Vn-tag) to the standard Ethernet frame to differentiate the different virtualized network interfaces (VIF) to identify the traffic of a particular virtual server. This scheme requires both the server and the switch to be modified at the same time, need to work together to identify and deal with vif. Another scenario is to modify the spanning tree protocol and reuse q-in-q to resolve the problem that the data frames specified in the STP cannot be sent to the port that received the frame, because this is allowed in a virtualized environment A layer of basic 802.1q markup is added to define the different virtual machines, which can differentiate different virtual machines by q-in-q outside the VLAN.

The "network" discussed above is the IP layer and low-level technology involved in cloud computing, and the TCP (Transmission Control Protocol) technology that provides reliability services for cloud computing will require some significant changes if the scope of the new technology is extended upwards. Cloud-oriented TCP has at least three problems, one is to shake the Internet core design concept, the second is to add a new type of communication mode for TCP, and the third is to improve congestion avoidance and control mechanism in TCP.

"End-to-end transparency" is the core design concept of TCP/IP, which assumes that the network is unreliable (IP protocol), so all the mechanisms of reliability (TCP protocol) are placed on the terminal. The basic assumption of this idea is that if the network is unreliable, then the communication terminal must be reliable, otherwise it can only interrupt communication services. Therefore, the problem of unreliable communication terminals is not considered within the traditional TCP/IP protocol. But from the perspective of cloud computing, if a cloud server (communication terminal) is down, the user should be allowed to automatically switch the cloud service to another server (from the TCP point of view, the server of different IP addresses) to ensure continuity and reliability of the service, but the TCP protocol is obviously not. The problem of service reliability in cloud computing depends on the application layer or the need to design a new TCP for cloud computing, which is still in the theoretical stage.

Secondly, we need to introduce the Incast topology model in the TCP protocol. Incast is a many-to-one mode of communication where a parent server initiates a request to a group of nodes (server clusters or storage clusters), and nodes in the cloud server cluster receive the request at the same time and respond almost simultaneously, many of which simultaneously to a single machine (parent server) Sends a TCP data stream, resulting in a "micro-burst flow". This occurs primarily in cloud computing data centers, especially those distributed storage and computing applications that are implemented in an outward-expanding manner (such as Hadoop, MapReduce, HDFs, etc.). At present, the data center TCP (dctcp) Technical Standard scheme has been proposed to deal with this problem.

Third, a growing number of people believe that cloud computing terminals also need to participate in cloud computing services, because there is no reason to use only the resources on the cloud server, waste the terminal on the increasingly rich and low-cost computing and storage resources. To solve this problem, it involves the negotiation of three kinds of resources between cloud terminal and cloud server, computing, storage and communication. Although the traditional TCP protocol also makes the communication flow rate negotiation between the server and the terminal, but there are two obvious disadvantages: 1 The traditional TCP resource capability negotiation is implicit, which is realized by conjecture, not explicit; 2 does not distinguish between computing resources and storage resources, and network resources in terms of economic differences.

Author Introduction:

He Baohong: Chairman of Internet Research field, Institute of Telecommunication, Ministry of Industry and Information Technology, director of Internet Center Vice-Chairman of IP and Multimedia Working Committee, China Communication Standardization Association, ITU-T SG16 (Multimedia), ITU-T Iptv-gsi and IETF domestic counterpart team leader