First, the concept of network architecture
Computer network is a set of computer systems connected by certain forms, it needs four elements of support, that is, communication lines and communications equipment, with independent functions of the computer, network software support, to achieve data communication and resource sharing. The computer network has two reference models, the OSI model and the TCP/IP model, in which the OSI model is the theoretical model, and the TCP/IP model has become the industry standard of the Internet fact, and now the communication network is generally using the TCP/IP protocol cluster. Application programming is programmed with socket sockets.
Second, the network layer model
Layering can make complex problems simple, and network layering is based on this principle. Network layering simplifies network design and improves the degree of standardization of network interconnection. Network hierarchy is the last layer is dependent on the next layer, only the bottom is the physical actual communication, the other peer layer is virtual communication. Layering principles and methods as shown, the network layering model involves some of the following terms.
① entities: The active element in each layer can be any hardware or software process that can send or receive information, and in many cases, the entity is a specific software module. ② Peer entity: An entity located in the same layer on different machines. ③ Service provider: N-tier (lower layer) is the service provider for the N+1 layer (upper level). ④ Service User: The N+1 layer (upper level) is an n-tier (lower level) service user. ⑤ Service access Point (SAP): The location provided by the service, which is the exchange of information between adjacent two-tier entities on the interface. ⑥ the relationship between services and agreements: services are vertical, and agreements are horizontal. N-tier service users can only see the N-tier service and cannot see the N-layer protocol, the N-Layer protocol control under the two peer-level communication allows n-tier to provide services to the N+1 layer, to implement N-layer protocol, the need to use the services provided by the N-1 layer.
Network layered Model DiagramBoth network models of the Internet (OSI model and TCP/IP model) are implemented based on the principle of layering. Through the network stratification, can obtain the benefit: Each layer is independent from each other, the adjacent layer interacts only through the interface, causes the entire problem complexity to descend. The structure can be split open, each layer can be used the most appropriate technology to achieve. Each layer features simple, easy to implement and maintain. A layer of changes, as long as the interface service does not change the relationship, the other layer is not affected, the flexibility is good. Stratification facilitates the standardization of network protocols.
Third, the specific objectives of the Internet architecture designThe core goal of network architecture design is "effective", but it does not show what kind of goal the effective interconnection should achieve. The following list sets out the detailed objectives of the network architecture design: 1) communication survivability in the face of failure, 2) support for multiple types of communication services;
3) Support different types of network access;
4) Support the distributed management of resources;
5) Consider cost-benefit factors;
6) The cost of host access to the Internet is low;
7) Support the billing statistics of network resources.
These specific goals appear to be a complete list of network system requirements. What is important is that we need to understand the importance of the sequencing of these goals. If the order changes, it will result in a completely different network architecture. For example, for military networks, which means that the data can be transmitted in a complex situation where the enemy is compromised, the ability to communicate survivability becomes the primary goal, and the billing management capability is the final goal. During the war, we were more concerned with how to reliably deliver the information we collected to our destinations as quickly as possible, regardless of the cost of communication. So initially people do not care about the resource billing management, and for the design of commercial network architecture, resource billing management should be placed in the first place. In the same vein, the cost-benefit factor is our goal, but it is the distributed management of resources and the support of different types of network access. Some popular commercial networks are optimized for specific transport media, such as a storage-and-forwarding network based on long-distance telephone lines that can communicate well at very low cost, but support for other types of networks may be poor, as commercial network architectures are designed to take more cost-effective issues into account.
Iv. Communication Survivability in the face of failure
Internet
The most important goalis a strong ability to survive, or to be self-recovery. If two entities communicate over the Internet during a failure that causes the communication to break temporarily, it can reconfigure the network Recovery Communication Service, and then the communication between entities can continue without having to re-establish a higher-level session. The Transport layer protocol does not always report a temporary network outage to the client, it always assumes that the failure can recover quickly, and then reports to the client after repeated attempts to determine that the network cannot be recovered. The choice of Internet architecture is to collect state information and save it at both ends of network communication. This approach is called "fate-sharing (Share life)". It means that one end of the communication disappears, and the other end also discards its saved state information. The status information of the transport layer is saved in the host, the host is connected to the entire network, and the service is used throughout the network. The "fate-sharing" mode has two advantages over the "copy" mode: First, it can not be affected by any network intermediate node failure; second, engineering is easier to implement. The "fate-sharing" mode means two points: first, neither the intermediate packet switching node nor the gateway holds state information for the current communication session, in other words, they are stateless packet exchanges. So this network architecture is often referred to as a datagram network. Second, in this architecture the role of the host is more important, because it does not like some network itself to provide reliable data transmission services, and it uses the host side of the transport layer algorithm to ensure that the data is ordered transmission, packet loss retransmission and confirmation, and the host application does not care about the details of data transmission.
v. Support for multiple types of communication services
Another core objective of the Internet architecture is the ability to support multiple types of communication services. Different types of communication services vary greatly in transmission speed, delay, and reliability. The most traditional type of transport service is a two-way reliable data transfer service. This service is also known as
"Virtual circuit"Services for applications such as remote login and file transfer, which is the primary service provided in the Internet architecture using TCP. It was early to realize that even traditional "virtual circuit" services did not meet the needs of diversity. For example, the remote login and file transfer, the former requires service delay can not be too large, but the bandwidth is not too high, and for file transfer, more concerned about the data throughput rather than transmission delay. Previously, TCP attempted to provide both services at the same time, and the results were unsatisfactory. Initially TCP was designed to be generic enough, but because of the diversity of the network environment, it was unlikely that a protocol would be used to carry it. So a better approach is to build a service that can handle every data over the network without requiring that every byte be delivered reliably and sequentially. TCP does not apply to the service is real-time communication transmission, conference calls and live TV use this service. The primary requirement for real-time communication transport services is that there is a smaller and smoother delay when transmitting packets, rather than requiring high reliability. That is, in the conference call transmitted voice in a timely and continuous line, do not need each syllable is clear and accurate. The approximate process is that the application layer will digitize the analog voice and package the binary data and send it sequentially over the network. These packets must be ordered in order to ensure that the correct voice signal is restored. If some packets do not arrive as expected, it is not possible to reproduce the voice in a timely manner. In the case of using TCP to do a series of research on the transmission delay control, finally surprised to find that transmission reliability is the main cause of delay. This is because a typical reliable transport protocol discovers that a retransmission request is sent when a packet is dropped, and the subsequent packet transmission is deferred until the packet retransmission succeeds before the sending window transmits subsequent packets. Once this happens, it can result in several times of additional delay, and may be completely interrupted. The resulting voice data is complete, but only intermittent and unrecognized sounds can be heard. This is obviously not appropriate, in fact, the handling of packet loss can be very simple. This is the use of an empty voice (no sound) substitution when encountering a packet loss, because in most cases this has no effect on the listener's resolution of the voice content, and then continues to transmit subsequent data. Even if it has an impact, you can either use advanced error correction, or the listener will let the other person repeat it all over again.
Support for multiple services in the Internet architecture should not be provided by the lower-level network, otherwise it would deviate from the goal of the Internet architecture design. Thus, we want the host and gateway to provide algorithms that abstract various services from the Basic datagram service (IP datagram).
vi. Summary and Prospect
TCP/IPThe protocol has been widely used in various fields and has derived a lot of similar architectures. The datagram is a good way to achieve most important design goals, but when we try to accomplish some of the following goals, we find that the management of billing statistics resources is difficult to implement in a datagram-based environment. Most
Data Reportis part of a series of packets that are transmitted from the source to the destination, and cannot distinguish between separate units at the application level. However, this series of packets is transparent to the gateway because the gateway processes each packet independently. Therefore, each packet must be independently billed and counted for resource management, due to the loss of Bao and other transparent to the gateway can not accurately calculate the actual amount of data transmitted by the application. In the network layer, the use of datagram model makes the network layer lose the important information source, and it is difficult to achieve such goals as billing statistics resource management. This suggests that there may be a better abstraction in the next generation of network architectures than datagrams. Regardless of any particular service type, it can identify a series of packets sent from the source to the destination, which can be described in terms of "flow". This method needs to remember the characteristic attributes of the data flow through the network, but it challenges the architecture principle that the network intermediary device does not record the communication state information. At the same time, as mentioned above, the distributed management control of the gateway and the Internet Protocol are becoming more and more bloated and complicated, so we can start to consider the centralized management control mode and the abstract encapsulation method based on "stream", which fundamentally challenges the Internet architecture. With that in mind, SDN and OpenFlow seem to be on the way to the evolution of the Internet architecture, and the separation of data and control surfaces provides technical support for a "stream" of abstract encapsulation, partly to subvert the centralized management control (network local) policy of distributed principles and to record "flow" based on Communication status information is convenient because cloud-based data centers can store and process flow state information efficiently and quickly, making the network smarter.
Attached: (1) TCP/IP model Http://zh.wikipedia.org/wiki/TCP/IP or Baidu
(2) Data Report significance reference: Design philosophy of Internet Architecture v1.0 (3) OSI model http://blog.csdn.net/yaopeng_2005/article/details/7064869
Understanding of the Internet architecture
