Technology disclosure: overview of data center virtualization
追风筝的人  2024-10-10 14:09   published in China

Source: Huawei data storage FusionCube hyper-converged infrastructure 2024 special issue


from virtualization to data center virtualization

so far, the virtualization we are talking about only focuses on the virtualization of computing resources, which can be called computing virtualization or server virtualization. However, virtualization is not only aimed at server resources.


From the perspective of a data center, virtualized objects should include computing, storage, and network resources. Computing virtualization has been discussed in the first section. Storage virtualization can be traced back to the 1980 s. By Professor Patterson of the University of California, Berkeley, RAID(Redundant Array of Independent Disks) was proposed in the paper "A Case of Redundant Array of Redundant Disks". Technology. RAID is originally a technology to improve reliability, that is, combining multiple independent physical hard disks in different ways to form a hard disk Group, in order to provide higher storage performance and data redundancy technology than a single hard disk, however, the RAID technology not only improves the reliability, but also realizes the abstraction of resources, users do not need to pay attention to which hard disk their data is stored, but the disk array provides a series of LUN(Logic Unit) for users to use. From physical hard disks to logical hard disks, RAID technology has blown the wind of virtualization into storage.

In traditional enterprise storage, one of the main functions of the so-called storage controller is to virtualize the disk array and provide LUN for the host. With the increase of data volume, especially the explosion of massive unstructured data, and the higher requirements for the capacity and scalability of storage systems, distributed storage emerges as the times require. Unlike traditional enterprise storage, distributed storage does not have a dedicated controller, but uses distributed software to combine hard disks in server nodes into a storage pool, and then uses virtual volumes to the host. The implementation principle is different, but the final result is that multiple hard disks are virtualized into a storage pool and provided to users as LUNs or volumes.


The successful distributed storage includes VMware's vSAN,EMC's Isilon and Huawei's OceanStor Pacific, while the open source has the famous Ceph. Due to its low threshold, it has gradually become the mainstream of storage virtualization, therefore, storage virtualization mentioned in many scenarios is equivalent to distributed storage, especially in hyper-converged scenarios.


Among the three major virtualization, Network Virtualization is a later technology and a more difficult technology to implement. In short, Network Virtualization refers to the separation of logical networks from the underlying physical networks. Early network virtualization mainly solved how to isolate and guarantee QoS in switching networks. VLAN, VPN, VPC, VPLS, etc. can all be classified as network virtualization technologies.


With the development of data centers, higher requirements are put forward for network virtualization: first, how to build a virtual network within servers, between different virtual machines, between containers and containers, to realize exchange, network functions such as isolation and forwarding. The second is that the servers form a cluster. How can the physical network in the cluster be separated from the logical network to meet the characteristics of multi-tenant access and on-demand service, while having high scalability. in response to these challenges, Professor Nick McKeown of Stanford University put forward the concept of OpenFlow in 2008 and published the classic paper OpenFlow:Enabling Innovation in Campus Networks. The Software- Defined Network (SDN) was boldly conceived, opening a new chapter in Network virtualization.


SDN, in short, is to establish a control plane on the network to uniformly manage and control the data forwarding of lower-level devices, while the management function of all lower-level nodes is stripped, leaving only the forwarding function. The network under SDN control becomes simpler. Managers only need to configure the network like configuring software to implement routing, forwarding, isolation and other functions. With the development of SDN technology, two genres have gradually formed. One is the classic-OpenFlow. OpenFlow has always had an ideal, hoping that the underlying switching network can be white-card, and the switch only provides a simple switching capability, the switching protocol and the North-South interface are all unified. All network control is completed by the SDN Controller. Even the SDN controller should be just a software that can run in this ideal. For the vast number of network equipment manufacturers, of course, it is fatal. Therefore, in 2013, with the support of the Linux Foundation, Cisco, together with IBM, Microsoft and other companies, launched the open source organization OpenDaylight. Unlike OpenFlow, OpenDaylight also agrees to have an SDN controller to control the network in a unified way, but this SDN controller should be a professional device, SDN control has better performance and reliability through professional hardware. At the same time, for the underlying switching network, it cannot be completely white, and should have some cooperative ability with SDN controller, In order to improve the efficiency of SDN network, it is also called hard SDN. Of course, this genre more reflects the wishes of device manufacturers, but in actual use, the complexity of the network determines that most customers still choose a relatively secure direction of hard SDN or a combination of hard and soft SDN, only a few enterprises such as Google that have advanced technical capabilities and relatively closed networks are really practicing the classic OpenFlow route. Computing virtualization, storage virtualization, and network virtualization together form data center virtualization. Virtualization technology has also moved from solving the problem of single server cluster resource management to solving the problem of data center resource management.

Replies(
Sort By   
Reply
Reply