Network Technology when the revolution

After the computer network has been born, the application scenario mainly includes the local area network, the wide area network two big categories. In various environments, the LAN is connected by a hierarchical structure, forming a network of networks, the so-called Internet. No matter what the network, the only fact of the standard is the TCP/IP protocol stack and the various management and application technology around this stack, even if the later launch of IPv6, CCN, SDN and other network technologies, are not completely beyond this category. So, it looks like TCP/IP-based tinkering has met the needs of the web for a long time in a variety of scenarios

However, in the current era of cloud computing, data center scenarios for the network needs, so that these traditional network architecture began to encounter real problems.

In the data, it has its distinct network requirements and operational characteristics:

• Resources are managed uniformly by one party (or a logical layer of unified queries)
• The network itself is highly dynamic (various migration schedules)
• The scale of the network is enormous (hundreds of thousands of hosts, even millions of hosts)
• Policy configuration is complex (a VM may follow more than 10 rules)
• Demanding performance requirements (not only for normal transmission performance, but also for convergent performance)

These characteristics of network technology put forward a very serious challenge, and even can be said that some of the problems in the existing network architecture, there is no solution.

For example, the problem of extensibility, the traditional two-tier aggregation, and then three layers of isolation broadcast domain practices, so that the network for the core layer of pressure surges. And even expensive commercial core solutions have limitations.

Again, such as the problem of migration, this fundamentally to the TCP/IP model design challenges, although there are various overlay technology to try to compensate for this flaw, but also bring the cost of transmission and additional management costs.

These challenges make network virtualization look beautiful, but the landing phase will find it very difficult.

So why not reconsider the whole question and see what the network needs in the data center are?

In the current data center, whether or not the virtual machine is actually running, the user is provided with a variety of Internet access services (the form of direct virtual machine to the user will be less and fewer). Users accessing these services are based on the TCP/IP service. Between these services, and how the internal implementation, in fact, is not necessarily TCP/IP.

Once you get out of the limits of TCP/IP technology, you can actually design a set of protocols that target a variety of scenarios within the data center. For example, in the data center, virtual machine (or host) discovery, in fact, do not need to rely on the traditional two-layer broadcast form. In the beginning of the virtual machine (or after the virtual machine boot), the control layer has been clearly known where the virtual machine, it needs to communicate with whom, then it is natural that the corresponding location information and communication information can actually be configured and optimized in advance, without having to broadcast the query interface, through a layer of even multi-layered response, Causing a lot of useless packets.

When the broadcast domain does not need to be isolated at all, even when the broadcast domain is eliminated, the pressure on the switching device will be much reduced. At that time, the design cost acceptable, performance to meet the requirements of the exchange layer will be possible.

Of course, not to completely abandon the traditional network in the field of all kinds of excellent technology and methods, a lot of ingenious ideas can be applied to the data center scene, so-called original aim. But there is no doubt that the protocol, which is not bound to TCP/IP, is not a reasonable scheme (and perhaps a solution that is not), which will inevitably lead to more problems.

TCP/IP protocol stack in its development process, has experienced too many waves, do not know this challenge, it can again safely through.

Reprint Please specify: http://blog.csdn.net/yeasy/article/details/42688499

Network Technology when the revolution