Evolution of data centers: abstraction layer, STP protocol, and TRILL Protocol

The multi-layer and multi-layer data center network age is almost eliminated. As the data center further implements various abstraction layer technologies, many enterprises have chosen to rethink how to design their data centers and migrate them to a flat simple network, this network can support more different traffic modes than in the past.

Abstract layer concept

You may want to know the meaning of the phrase "abstract technology" mentioned in this article. Most of these technologies are centered on various Virtualization Technologies, which can be purchased on the market.

Server Virtualization

The rise of virtual monitor hypervisor is the first development, which has brought us today. The data center today seems completely different from the data center 10 years ago. Although previous data centers often had groups of operators processing physical tasks, such as adding new hardware, deploying new operating systems, and installing new cable systems, however, these tasks do not happen as often as they do recently. Of course, some enterprises still need to deploy new hardware, but from an operational perspective, service deployment has been migrated from hardware-intensive tasks to major software-driven tasks. At the same time, the workload of each physical server is also increasing, thus reducing the number of physical servers that need to run the workload.

However, the abstraction of this workload also puts pressure on related data center resources, including storage and the overall input and output ports.

Storage Virtualization

For many enterprises, the era of massive storage of a single server to meet the needs of applications no longer exists. Instead, it is a new era of large-scale shared storage, which is combined with many different tandem storage layer operations to balance workload and costs. To prevent storage confusion, some commands are derived. Currently, we can see the abstract storage management responsibilities, efforts are made to increase the level of a single array and a higher level away from the pool storage including the entire system. When an enterprise looks at storage virtualization from a higher level, it looks like server virtualization. A single workload is separated from the underlying hardware, while a complex software system determines the affiliation of a workload.

Input/output port Virtualization

How many people need to carefully hide a large number of cables in your server cabinet? You may have installed cables to connect servers to many different data networks in the future, connect additional cables to the storage system-whether they are iSCSI commands or fibre channel-or many other types of connections. Network adapters and switch structures that are rapidly emerging at an astonishing speed can handle a large amount of workloads, as we can see, the emergence of these suppliers can be combined with a completely different connection method to a single cable, to manipulate the server that can carry all the necessary traffic and support the workload required by the server. Xsigo and other companies claim that they can reduce the number of required cables, interface cards, and switch ports in a complex data center environment by up to 70%.

Continue to use Xsigo as an example. Their solutions are migrated to a software layer and there is a need for most work to support server communication, administrators can manage data center networks with unprecedented time limits and flexibility. In a real case, we were engaged in System Engineering 12 years ago. We had to run six independent network cables to each of the three fully-architecture servers, connect each cable to the network switch port. Now, we can achieve the same goal with only one single cable, A 10-gbe connection, and a switch port, and cut it into any way we want to connect to the 10-gbe. What is this concept.

Profound impact

All of these abstract technical means can improve the efficiency and availability of various new opportunities, and significantly change the way traffic is run in the data center. Communication traffic was once a general integration and outflow data center, and internal traffic in the data center increased exponentially. Although applications always need to communicate with each other and different application layers sometimes have to chat online, when you start to consider running a large amount of work load on a single host, things change as they move any workload that runs in the data center. Simply put, tools are becoming increasingly bandwidth-intensive, which was unpredictable 10 years ago.

Spanning Tree Protocol

Over the years, researchers have tried to control network problems by using the Spanning Tree Protocol (SPT), which usually works well, despite the successive challenges posed by many new protocols. But the SPT protocol has always been the most advantageous.

However, the SPT protocol is based on link blocking rather than trying to make the most effective use of available network resources. Recently, high bandwidth needs to be applied inside the data center. However, such a method needs to solve critical network problems, which may lead to lower-than-optimal performance, it may even meet the traffic growth at the cost of increasing costs.

The spanning tree blocks a port.

TRILL)

If we don't worry about Loop Blocking links, we can constantly use any Controllable network path. In this way, the network path utilization is not reduced simply by a loop.

This is a newly emerging technology called TRILL-transparent exchange-beginning to play a role. Simply put, TRILL will make you use all the network paths with no scrubs at all times. It does not actually need to generate a tree environment.

By using TRIL technology, all paths form a large network, where all paths are valid. By keeping all paths available at any time, the data center can better support high-bandwidth and Low-latency workloads.

TRIL makes all connections available

Debate followed

As we can imagine, throwing away a technology that has been well used for years does not apply to everyone. In the network, the question is whether to adjust the traffic burst mode of the data center.

Supporters of the Spanning Tree Protocol said the Protocol has existed for a long time and companies need to be cautious before using the new protocol. The protocol can be generated or canceled, but it is rare for a protocol to span from an earlier LAN to a 10-gbe network today.

The other side of the debate is those who feel that each protocol has its value, and that the spanning tree will always disappear in favor of more flexible and effective protocols. Another type of users advocate a combination of two protocols to replace TRILL's core network, and add smaller STP Domains Based on the grid core to reduce the impact of downtime.

Reduce network level

Finally, let's talk about the three required network layers. Traditionally, we already know the three layers of network design in our minds:

Core Layer

Distribution layer

Access layer

These three layers are designed together with performance, security, and scalability. In addition, it allows the network technology to be deployed, which may not combine multiple layers into one unit. In a dual-layer network, the distribution layer is canceled, and its services are accumulated into cores or decomposed into access layers. This is one of the most common practices for processing these new clustering layers with modern devices that can surpass and maintain the consistency with the vendor's sales devices. Two-layer networks are easier to maintain redundancy and have the same scalability performance as three-layer networks.

Summary

As the times change, today's data centers are constantly making adjustments to adapt to the new challenges they face. Therefore, as a data center manager, it is necessary to consider the overall architecture of the data center based on the constantly changing traffic and operation modes.