Pain points of data center and Storage Network merger

Pain points of data center and Storage Network merger

The data center network has always been two parts of the world, with both the data network and the storage network accounting for half. The data network uses the Ethernet IP technology, while the storage network uses the FC fiber channel technology. For the historical reasons why there are two types of networks and two technologies, you can ask du Niang. I don't have to elaborate on them here, but from the day when the data center was born, we will discuss how we can combine the two networks into one. The benefits of combining the two networks into one network are obvious: cost saving and no need to build two networks, reduce the number of network interface cards required by the server, reduce the cost budget of the server, and reduce the cooling requirements. greatly reduce the number of cables to be managed for easy maintenance, without the need to master Ethernet and Fiber Channel technologies, you only need to master one. However, the reality is cruel. After all, the two network technologies are too different. They each have their own characteristics and cannot be replaced by each other. The principles of data forwarding and the principles of data storage are completely different, this is like a server also has a hard disk or can store data. However, after all, a storage device is deployed without the capacity and data fault tolerance mechanisms of the storage device, so the two cannot replace each other. Since the devices at both ends of the network cannot be replaced, can the devices in the middle of the network replace each other? The answer is yes. Both the data transmission network and the data transmission and storage network define the Message Protocol format. The IP protocol format and the FC protocol format are naturally different, so they cannot be used by each other, but they can be encapsulated, the outer layer of the FC protocol is encapsulated by the IP protocol, or the outer layer of the IP protocol is encapsulated by the FC protocol, so that data can be transmitted over the Ethernet network or transmitted over the Ethernet. We know that the Ethernet IP technology is developing rapidly, which leads to the increasingly mature Ethernet IP technology. In contrast, the FC technology is developing slowly. Under such circumstances, it is expected that the data network will be used to transmit and store data, so that the two networks can be merged. Therefore, a new technology-FCOE technology was born.

FCOE is the technical basis for the integration of the two networks. In essence, it adds an FCOE packet header outside the FC packet to realize the purpose of transmitting and storing packets over Ethernet. By deploying a FCOE network in a data center, you can use an Ethernet network to complete data forwarding and storage. This technology is a blue ray for FC network equipment vendors. Once this technology is mature and widely used, FC network equipment vendors will have to close their doors. Obviously, this process will not be smooth, and the data center has to go through many pain points to merge the two networks. First, the storage traffic is very sensitive to packet loss and latency, but this is not the case in the Ethernet. However, the storage traffic transmitted in the Ethernet requires zero packet loss and low latency, this puts forward high requirements on Ethernet technology. With the popularization of 10G/40G network technology, high bandwidth solves the latency problem, and new data center network devices also have large cache technology, in addition, the latency of Data Forwarding is still low in the case of congestion caused by burst traffic. A separate traffic control technology is designed for FCOE to avoid packet loss in the network. When the server or storage sends a traffic control frame to the network device, the network device can transmit the traffic control frame to the sender, which requires a speed reduction. If the congestion condition is not serious, the network device can also cache the traffic in its own device. The Flow Control Technology in FCOE is different from the flow control technology in Ethernet, and the flow control effect is better. Although these technologies are supported, they still encounter difficulties in implementation, we will find that there are many traffic control frames in the FCOE network, which leads to unstable traffic in the FCOE network, but always traffic spikes. This is caused by frequent traffic control, if the storage volume is large, it is a test for the FCOE network. Second, the traditional L2 Ethernet uses the Spanning Tree Protocol to maintain the loop elimination. This means that a redundant cascade switch can only use one of the cascade bandwidths, which is a waste of bandwidth. At the same time, the Spanning Tree Protocol may cause too long paths, low switching efficiency, and long switching time. In this scenario, it cannot meet the requirements of storage traffic forwarding. That is to say, if the FCOE network is deployed, the traditional STP, MSTP, RRPP, and other ring network protocols cannot be used. The new ring network protocols, such as TRILL, Fabric Path, and SPB, must be used, these new ring network protocols can theoretically solve the problem of low redundancy links and low link switching efficiency, but the actual results are still unknown. It is understood that the deployment of TRILL and other ring network technologies in the actual data center is still very rare, and a long process of acceptance is required for the new technology. Third, there is a problem of personnel and process changes. Storage technicians do not want to work for Ethernet technicians. They know that there must be winners and losers, and no one wants to be the loser. The "Brute Force" across different departments in the data center also plays an important role in Ethernet and storage network transmission. Finally, the FCOE technology is not easy to maintain. Although FCOE is an Ethernet-based encapsulation technology, it is actually an FC packet. When a problem occurs, it is very troublesome to locate and analyze the problem, you need to analyze the problem at the same time by combining servers, networks, and storage. In particular, the differences in the implementation of FCOE in the server, network, and storage fields may cause such problems when devices of different vendors work together, the problem cannot be solved. Once such a problem occurs, the damage to the customers in the data center is serious. Many customers in the data center often switch to the FC network, or simply separate the data network from the storage network, FCOE technology is often removed immediately. This is because the FCOE technology itself is not very mature. It also requires personnel who know the IP and FC technologies and the tacit cooperation between devices of different manufacturers, FCOE technology is very popular for various reasons, but there are not many data centers actually used.

Putting the data traffic and storage traffic in the data center on the same network is equivalent to putting the highway and railway traffic together. This is feasible, however, we need to be especially careful in preparing and monitoring, and pay attention to the many pain points faced by FCOE technology. These pain points have not been well resolved so far, this also makes the integration of the two networks of the data center an empty talk, making it difficult to integrate the data center solution.