Datacenter migration, data center is now underway, with ever on server performance and scale, in order to bring more server access to the data center, the scale of network is becoming more and more big, in order to connect all the devices, network become intertwined and interconnected relations is extremely complex, network part management is becoming more and more.In such a development context, it is necessary to point out a correct development path for the development of data center network, otherwise the network seriously restricts the development of data center.How to keep the network complexity constant or even decrease in the case of increasing the scale of data center construction?This has become the most active topic in the data center, and it is the biggest demand of data center in the development process. What should we do?
First of all, the traditional data center network construction is a three layer architecture: access layer, convergence layer and core layer, is designed so that, in fact many data centres in the presence of a multi-level united access part, from the core to end server, often after several network devices, because the network equipment of port and capacity is limited, the advantage is can greatly increase the number of network can access devices, but too much network level, bad management is conducive to business flow forwarding, potential failure, too much to make radical changes from the network architecture.Now the network equipment in the performance and scale is larger, a core network equipment can often provide thousands of Internet access port, the hundreds of thousands of sets of server equipment, so the data center should try to use the larger network equipment, network architecture is composed of three layer to the second floor, ignore the convergence layer and at the same time try to reduce the number of cascading access layer.On physical connection, maintain export from the server to the core network equipment, only through the network access layer devices, such forwarding efficiency greatly, the network structure is also becoming clear, when the need for business expansion, transverse expansion directly, rather than the longitudinal expansion, make network forwarding plane is becoming more and more wide, rather than getting longer and longer.
Secondly, the network physical level is reduced, and at the same time, it is necessary to make changes at the same time.Traditional network is two or three layers of mixed running, access layer basic do second floor processing, to gather all three layers of interconnection and core, data center also keep three layer connected with the outside world, this data center network main body is three layer network.The three-layer network greatly restricts the development scale of the virtual machine network, and the virtual machine needs to be in a two-level communication network to complete the smooth migration of the business.Virtual machine is one of the necessary conditions for rich data center service deployment, cannot meet the virtual machine migration will be restricted to cloud computing data center, big data fields, however, more data center at the time of construction is in accordance with the three layers of network design, need completely over to change for the second equipment, it is difficult to implement, and external data center may not is the second floor, backbone all three layers of interconnection, and migration between virtual confidential data center outside, must be in other ways to find a way to, so big technology came out on the second floor.The second layer technology is designed to build a virtual two-layer network on the three-layer physical network to meet the demand of virtual machine migration.The entire data center, even the data center, is in a two-layer logic network, which can not only satisfy the virtual machine deployment business, but also the network structure is very simple and easy to manage.
Third, network equipment to be virtualized, with the upgrading of network scale, the equipment quantity more and more, with hundreds of thousands of network devices is not surprising, to manage to so many devices at the same time, difficult.From an operational perspective, the need to introduce virtualization management, and network equipment is virtual, virtualization as a device, several sets of equipment to manage virtual devices, so that greatly reduce the number of equipment management, at the same time, you can ignore these devices interconnected relationship between number of equipment as a whole to provide services to foreign.Common core equipment can support 2 ~ 4 platform virtualization, access devices often can reach more than 10 sets, can even core layer and access layer of equipment all do virtualization, make the whole network equipment have been formed between virtualization.The data center operation and maintenance personnel are looking at the virtual several network devices, the interconnection and the port is simple, very good management.This also greatly increases the redundancy of the network and improves network stability.
Finally, business deployment automation.When traditional data centers are adding servers or businesses, they also have to make configuration changes on the network, which is tedious and error-prone.Now the controller to do these tasks to network business deployment of issued by the controller, when a new server access networks, automatically finish the server operating system, download, install, and part of network configuration, all are done by the controller, now just need to server access to network devices, connected UP UP.The controller can not only complete the new business on-line, but also can conduct network business deployment and operation and maintenance.Forward flow table is issued by the controller network, to guide the entire network data forwarding, the network control layer and the data forwarding plane separation technology has been widely recognised in the data center, when faced with failure, the controller can automatically business separation, automatic recovery.Of course, these seemingly simple processes require very complex designs for the controller, considering the various anomalies that might occur in the network, and deploying them in advance on the controller.The controller is a machine that can take the place of people, according to people's thinking, how to do it or to ask people to design well in advance.
The central idea of simplifying the network is to create a virtual two-layer network based on the existing physical network, thus greatly simplifying the work of all aspects of the network., of course, in the face of all personnel data center will be a virtual network, once a failure occurs, screening effort, because the virtual network found on the fault, to find the location of the specific physical device, also need to know the corresponding relationship between the two, even a virtual network node corresponding to a number of physical nodes, can controller, do the records of these things, when you need to know the corresponding relationship, directly out of the controller.Does not reduce network physical connection relationship are the main network from "Jane" method, sometimes with the expanding of network size, the physical connection to increase is inevitable, mainly in the virtual network level fluctuation kongfu, construct a simple set of virtual network.
沒有留言:
張貼留言