Sdn和传统网络的区别

《Sdn和传统网络的区别》由会员分享，可在线阅读，更多相关《Sdn和传统网络的区别（11页珍藏版）》请在装配图网上搜索。

1、SDN and traditional network the main difference lies in their different network architectures. In traditional network architecture diagram, the most important thing is to control layer and data layer separation. Each level has different tasks, layer with layer provides the data forwarding, routing f

2、unctions. Here, the control layer is responsible for the equipment configuration of the routing and data flow procedures. When you manage a switch, you are actually in the deal and switches control layer. Like a routing table, spanning tree protocol and all these things are calculated by the control

3、 layer. These tables built from such as BPDU (bridge protocol data unit, used to run the STP switches to exchange information between true), the Hello message such as frame relay, according to these news frame, switches to determine the available forward path. Once the packet forwarding path, the pa

4、th information will be sent to the data layer down, usually stored on hardware. Data level usually choose the latest by the control level for message forwarding path information transmission to come over. This model is very efficient in traditionally, the decision-making process of hardware is very

5、fast, the overall delay controllable and control plane can handle heavy configuration requirements. There are no problems with this approach, we focus on scalability. In order to prove the scalability problem, with our quality of service (QoS) as an example. QoS allowed according to the characterist

6、ics of the frame, according to the requirements of the scheduling, priority forward specific data frames. This to some extent reduced the specific traffic congestion in the network data transmission delay. Delay-sensitive, for example, voice and video traffic is classified as high priority and forwa

7、rded to ensure that the user experience. Traffic priority is usually based on the level of service (CoS) of a data frame or distinguish service code point (DSCP) tag. The frame must be unified in the data frame into the network, then the corresponding rules must also be set in the network, the deman

8、d in the traditional multiple exchange network becomes awkward, because each device needs to have the same configuration information. To illustrate the current network management challenges, we consider that each port on each device node in the network, the administrator needs to be configured indiv

9、idually, such work is very time-consuming and error-prone and awkward. In addition, in the data classification and appropriate routing network challenges still exist. For example, now we have two kinds of completely different data traffic, is a kind of iSCSI traffic, is a kind of voice traffic. ISCS

10、I as the storage flow, usually packets are full size, and sometimes there will be a huge data frames; While voice traffic is usually in a small packet transmission. In addition, there are different two kinds of traffic transport demand: voice traffic is sensitive to delay, this is to ensure the qual

11、ity of voice communication, the iSCSI is sensitive to low latency, but need more bandwidth. Almost without any tools in the traditional networks can differentiate between the two kinds of flow path and choose different depending on the type of traffic data to meet the specific needs of two kinds of

12、traffic. Is SDN hope to solve all these problems.1. SDN architecture According to the definition of ONF, SDN is divided into infrastructure layer, control layer and application layer, as shown in figure 1. Virtualization in infrastructure and control layer on two levels, the equipment level of virtu

13、alization, such as a physical support multiple logical switch; Which realizes the network level virtualization, first is SDN controller will of the entire network as a logical super switches on management control, the second will be the physical resources further according to the port, the media acc

14、ess control (MAC) address, IP address and other information is divided into multiple virtual network in accordance with traditional practice in the field of communication, in the architecture diagram below for south, above for the north, so the interface between infrastructure and forward layer call

15、ed south interface. ONF standardized is OpenFlow protocol, the Internet engineering task force (IETF) routing system interface (rs) protocol is being worked out. Control layer and application layer called north to interfaces, the interfaces between the industry mainstream implementation is based on

16、the hypertext transfer protocol (HTTP) RESTful interface, the concrete programming interface differ according to the different application scenarios. Figure 1 SDN layered architecture enlarge images In a more generalized SDN architecture, control layer and business choreography layer, the main resou

17、rces of SDN domain between the unity of the unified management, SDN network and other resources scheduling, such as 0 penstack + SDN data center solutions. Unified dispatching calculation, network and storage resources, it is equivalent to the business choreography layer of SDN. Standing in the poin

18、t of view of SDN, how control layer is divided into the concrete behavior of vendor application solutions, implementation, as the transmission control protocol, network protocol (TCP/IP) dont care about the application layer further layered design, referred to as the application layer. Standing in t

19、he whole network architecture level SDN, industry exist different opinions: (1) SDN only regional network renovation, to SDN control domain as a super equipment. SDN transverse interface does not change the original network, border gateway protocol (BGP)/multi-protocol label switching (MPLS) is stil

20、l valid. (2) SDN control field definition specifically/enhanced SDN east-west between interfaces, SDN as the entire network control plane. The author believes that the first scheme is more realistic, conducive to the smooth evolution of the network. The second solution of east-west interface can eit

21、her through the expansion of existing BGP, MPLS protocol implementation, or can be realized through the north to the interface in the aspect of business choreography, if you want to define more specialized SDN east-west interface, although it is possible to enhance the ability of the whole network,

22、but also increase the difficulty for deployment, the industry is under exploration. 2. The ZENIC architecture and key technology to realize control surfaces Implementation is based on the existing open source from academia SDN controller OpenFlow agreement, the forward model is also bound to a speci

23、fic OpenFlow protocol version . For the commercial system, must consider the entire product life cycle agreement the compatibility of the interface, consider the difference of different application scenarios and more manufacturers, the difference of multi-protocol interface, therefore SDN control su

24、rfaces must be set a compatible version OpenFlow, a variety of forward control protocol and the different ability of abstraction, we call forwarding abstraction layer (FAL), on top of this for the network operating system (NOS) core and the application layer provides the interface is independent of

25、the specific agreement and the ability of hardware. In OpenDaylight, this level is called a business abstraction layer (SAL) . This paper implemented a SDN controller - ZENIC, its architecture is shown in figure 2. Figure 2 top-down mainly includes protocol stack, driving and forward abstraction lay

26、er, NOS kernel and application layer. Figure 2 ZENIC architecture enlarge images 2.1 forward abstraction layer and drive layer Forward forward abstraction layer defines a unified control interface, including the abstract forwarding state below, turning ability, hardware resources, published, read/op

27、eration such as statistics, equivalent to drive the base class. Forward abstraction layer also forward management face driver instance, according to the forwarding plane when access to the basic ability to negotiate the different instances of drive, will forward the control connection is bound to th

28、e corresponding driver instance. Each specific device driver implementation forward abstraction layer interface, complete different interface protocols and hardware ability to forward the unification of the abstraction layer adaptation. From the point of view of control surface and the upper applica

29、tions, FAL transmit manipulation interface provides a unified, but due to the forwarding the capacity difference is bigger, the application for forwarding the operation there is the possibility of failure, therefore FAL need to provide application forward interface surface ability get/negotiation. Z

30、ENIC is implemented for OpenFlow1.1 adaptive negotiation / 1.2/1.3. 2.2 the network operating system kernel layer NOS kernel layer is the management of the network, the system resources, including topology management, host, interfaces, resource management, publication management, and manage the phys

31、ical topology, virtual topology, turn in a network of information database, etc. In general, the kernel layer, there is no default forward network logic to handle, but to preserve the accurate network topology, the resources status and storage, synthesis of the published, accept the application for

32、subscription and applications of network, resource state for network resources, forward logical operation. Topology management, the implementation of the current can be implemented based on standardization of OpenFlow cycle distributed across the link detection is based on controller message, Ethern

33、et is generally based on link layer discovery protocol (LLDP) implementation. Forward this implementation has the advantage of surface completely without perception, the disadvantage is that more link and shorter test timer, controller of high overhead. Another way is to have the forwarding plane ma

34、intenance link test timer, to detect, report will state that the implementation has the advantage of control surface overhead is small, the disadvantage is that need to be forwarded surface have certain default logic. The kernel layer is not only to maintain the network nodes, topology status, but a

35、lso need to collect the basic host location, status, which can be applied to provide a complete network view, further make forwarding, business decisions. Network virtualization should be built-in support for SDN controller. Should be built-in support for virtualization. Virtualization is the forwar

36、ding plane resources first division and isolation, such as according to the ports, logic, the host MAC address and IP address section for the division of the virtual network, the second is the virtual topology for customer/application permissions management. OpenFlow flow table model as well as for

37、switches, flattening management unified view has brought about many problems, including switching hardware complexity, not flexible, host, and to be tightly coupled. in the ZENIC system, inline network management as one of the kernel services, decoupling access networks and the Internet. The kernel

38、management of Internet network encapsulation format, upper application need only decision SDN control domain two access port position and strategy. The kernel to calculate the complete end-to-end path, and then forwarding decision by access side is mapped to the interconnection network path packagin

39、g labels. ZENIC supports a variety of Internet encapsulation format, including MPLS, virtual local area network (VLAN), etc., the next step is to support the virtual local area network (LAN) extension (VXLAN)/generic routing encapsulation protocol (GRE). This mode of access to the Internet, the appl

40、ication of completely without awareness, focusing on the host access side strategy. At the same time within the network management itself also can open interface, support custom routing algorithm and strategy. 2.3 north to application programming interface North to application programming interface

41、(API) in the different application requirements in the scene is different, also have to the requirement of packaging. If the network ability of atom exposed to the application, it is possible to form a unified API, but due to lack of encapsulation and ease of use, application programming, implementa

42、tion complexity is higher. Such as manufacturers realize the equipment level of open API up to more than 700, covering almost all protocols and equipment features, but for SDN, there will be at least two types of applications, different requirements: (1) professional network applications Customized

43、specification is high, need more details of the API, to the operation of the underlying network control ability is strong, such as routing protocols, custom tailored development intensification of traffic scheduling. (2) the common application The network as a service, just request network to provid

44、e service for application, dont care about the network details. In the latter case, north to interface to encapsulate A few best model and interactive service interface is simple, and easy to understand, such as to create A network request from switches A port to the switch port 2 B A l lGb/s bandwi

45、dth guarantee access, rather than by the application turns published and distributed to the path switches individually corresponding queue configuration parameters. There is a north to the ideas of the interface is defined by the application itself to the demand of the network and operation interfac

46、e, network vendors plugin to realize the application of network interface. Typically it is Quantum components, it defines the network API, provided by the various manufacturers Quantum plug-in - to control In own SDN controller or network devices. This architecture is equivalent to the SDN north int

47、erface standardization work up to the application, network adapter application requirements. Both advantages and disadvantages of each train of thought in north interface defined by SDN is idealized, trying to solve all problems, but its not possible for the network to understand the application req

48、uirements, standardization of advancing the work is relatively difficult, but also it is difficult to guarantee ease of use; Application driven model facilitates the SDN landing, but exchange between applications and multivendor network to a greater cost. ZENIC provides basic fine granularity of the

49、 underlying API, while providing encapsulation of API, virtual network has provided it is Quantum plug-in - In access to it. 2.4 distributed processing algorithm The distributed architecture of control surfaces and SDN separation architecture brought forward control state synchronization overhead, a

50、ccurate SDN global view can ensure the accuracy and real time of decision, for a applications such as load balancing can improve resource utilization, but need more frequent information synchronization, which greatly reduces the performance of the system. Starting from the design USES two kinds of m

51、ethods: controller is distributed as far as possible reduce the message copy; Control forwarding state synchronization between configured by the user according to the demand, necessary and sufficient condition only copy. Traditional cluster network system control surface is basically based on the di

52、stributed processing process, such as different business process distribution on different cpus, but a kind of process is still a single instance or a few instances, the parallelism is limited. In a single business process under the condition of sudden load, such as autonomous domain the way by adju

53、sting the border gateway protocol (BGP) process is the bottleneck. For SDN, as a result of the control network could be far higher than that of the cluster router, node number of the control surface of abortion is more demanding, so this method is the bottleneck is not feasible. Distributed hash tab

54、le (DHT) algorithm provides a scalable distributed data storage/routing algorithm. For the traditional application of unstable network Log2 (N) to find the complexity of the algorithm, the data center, telecommunications network applications, the industry a variety of enhanced one hop algorithm is p

55、roposed, based in part on a single hop DHT enhanced structured query language (SQL) No - open source systems have also been through commercial test, including the chateau marmont, Cassandra, etc., the first open distributed algorithm adopts DHT SDN controller is onix feeds, OpenDaylight project in t

56、he near future are also mentioned by Cassandra as the underlying distributed database system. The authors team also realized the improved single hop DHT algorithm . DHT algorithm based on consistent hashing, apply to a Key Value (Key, Value) storage model, type of structured query language (SQL) sup

57、port need to be further encapsulation. For SDN controller, the topology information is global, not suitable for DHT storage, but the need for additional global replication. Forward equipment related information organization in exchange for a node as a unit for distributed storage, can satisfy the ba

58、sic requirement, but granularity coarser, unfavorable to the load balance between the control node. Can host information by IP address, MAC table two dimension distribution, more even. 3. The forwarding plane forward abstraction technology OpenFlow 1.0 provides a single abstract model of the flow ta

59、ble 91, OpenFlow after 1.1 defines a model of a multistage flow table. 12 rs and parts manufacturers open interface to the application of exposure is a routing information base (RIB) on a variety of business table, the table an implied agreement between all kinds of logic. OpenFlow gave application/

60、control in the face of forwarding plane manipulation ability, to a great extent in theory can not be restricted by the existing network protocol completely, forwarding plane can be completely making a fool of instruction execution engine, and other open API is open API, under the framework of existi

61、ng agreements have strict limit condition. OpenFlowl. 0 is very simple, but need to three states content addressable memory (TCAM) support, and the price of TCAM is relatively expensive, at the same time the single table structure makes forward complex logical decomposition is very difficult. In the

62、 existing based on application-specific integrated circuit chip (ASIC) on the switch of OpenFlow1.0 above can easily be mapped to AcL lines, thus support OpenFlow Ethernet switches on the market at present the vast majority are only support OpenFlow 1.0. OpenFlow 1 x provides a multi-stage flow tabl

63、e model, added more table matching fields and instruction type, ability is more and more strong, but far from existing switches ASICs ability more and more. Software switch can easily realize OpenFlow1. X more table model. Hardware switches can through their own traditional ASIC assembly line for so

64、me necessary encapsulation, the formation of multistage flow chart to control surface, adapted by the control surfaces, only support instructions issued by the forwarding plane and table structure. This increase in counter rotating and controller are put forward higher requirements. Industry there a

65、re a few manufacturers launched a configurable ASIC link TCAM running water, these will be a fixed width of TCAM look-up table processing unit into smaller shard, such as every 32 bit TCAM is a basic fragmentation. Flexible application can define multiple subdivision level OpenFlow flow table, which support the OpenFlow multistage flow table model. Applications can exchange of L2, L3 switching, routing, such as the ACL decomposed to different on the flow chart of implementation, thereby avoiding the super-long top table keyword unnecessary TCAM costs.