Cloud Fabric—Creating a Future for Cloud Networks

Innovations must be made in network architecture and core switch hardware to support capacity and function expansion on cloud networks

Rapidly developing cloud computing applications have brought great changes to servers and storage devices in datacenters. As a result, datacenters must change their network architecture to adapt to these new cloud computing applications. The solution is cloud fabric architecture that meets the challenges in the cloud computing era.

Bottlenecks in Cloud Network Development

It is well recognized that enterprise services will migrate to cloud networks in the future as new cloud computing technologies develop and become more widely used. Enterprise CTOs/CIOs are increasingly concerned with what they should do to adequately prepare a traditional network to migrate to a cloud network.

We discuss here the three major changes that a datacenter network will encounter in terms of capex and opex during migration to a cloud network. If datacenter networks cannot cope with these changes, capex and opex will increase dramatically, making networks unsustainable.

#1 Networks are frequently upgraded because of fast changes to servers and storage devices, resulting in a sharp increase in capex

In accordance with Moore’s Law, IEEE and Intel predicted that server interface speed will increase more quickly in the next 10 years.

10GE server interfaces have been used for over 10 years since 2000, and in the next decade, 10GE/40GE/100GE interfaces will coexist on datacenter networks. If a datacenter network cannot support all four generations of servers, the network will have to be upgraded frequently when new servers are deployed.

Additionally, new technologies used on servers, such as IPv4-to-IPv6 and virtual machines, also challenge the stability of network architecture.

Centralized data storage and the explosive increase in unstructured data also bring great challenges to storage networks. As predicted by IDC, data to be saved will increase 44 times from 0.8 ZB (1 ZB = 109 TB) to 35 ZB between 2009 and 2020. Such a large amount of data will be supported by new technologies, including FCoE, lossless ethernet, and incast traffic model in distributed storage network. These technologies may cause changes in network architecture if the network architecture is not scientifically designed.

Therefore, a network design should be able to anticipate future changes and avoid rigid network architectures to control capex increases caused by frequent network upgrades.

#2 Network virtualization and dynamic VM migration bring changes to network management, resulting in sharp increases in opex

In the cloud computing era, datacenter networks will change from ‘static’ to ‘dynamic’.

In traditional datacenters, servers are fixed and require only static network configurations, making network maintenance simple. In cloud computing datacenters, virtual machines (VMs) are configured on physical servers and can migrate dynamically in a wide range. This requires on-demand, real-time, dynamic changes in network configuration, which makes network maintenance much more difficult.

Cloud computing transforms physical networks into virtual networks.

Many tenants (services or departments) use cloud services on the same physical network. To isolate different services or departments, the physical network needs to be virtualized into multiple logical networks in an on-demand, real-time, and dynamic manner.

If a complex network architecture is used, maintenance personnel must manually modify configurations for migrating VMs, and this greatly increases opex.

#3 Fast changes in network use require next-generation networks, resulting in sharp increases in both capex and opex.

Traditional networks are closed and separate the network layer from the application layer. Administrators control any interaction between the network and applications. In the cloud computing era, this closed model will reduce service deployment efficiency and response speed, resulting in lower customer satisfaction and a network that cannot meet the requirement for real-time response.

If networks are not open to peripheral systems, enterprises must construct new networks that can adapt to fast changes brought by cloud applications. This will result in sharp increases in capex and opex.

Therefore, the three roadblocks in cloud computing development are rigid networks, complex networks, and closed networks.

Cloud Fabric—the Best Choice

Cloud fabric, a next-generation datacenter network architecture was developed to help customers cope with problems faced by cloud networks. Cloud fabric is a revolutionary datacenter network architecture with the following key features:

Elastic Network

Innovations must be made in network architecture and core switch hardware to support capacity and function expansion on cloud networks.

#1 Innovation in Network Architecture: A new network architecture is required to allow flexible service deployment.

In traditional network architecture, the edge layer and core layer have similar functions. If you deploy cloud services in this architecture, the entire network must be frequently upgraded when you change the configurations of servers and storage devices to support new features. This makes network maintenance difficult. Moreover, network performance degrades when you deploy many complicated services on the same device.

Learning from successful practices of the Internet, an elastic network architecture has been developed that moves complicated applications to network edges to simplify configuration of the core network. The architecture has a service control layer that separates variable services from network devices. Therefore, this architecture ensures high network performance while meeting requirements of complicated and frequently changing services. When enterprises need to deploy new services, they can upgrade the service control layer without changing network device configurations.

#2 Innovations in Engineering Designs: Core switches with super high performance need to be manufactured to support four generations of servers (GE/10GE/40GE/100GE). The switching capacities, heat dissipation capacities, high-density ports, and high-performance chips of core switches must all be designed delicately to ensure that the switch capacity is sufficient for servers deployed in the future.

Easy-to-Use Network

On-demand, real-time, dynamic management of cloud networks requires the following innovations:

#1 Automated Network Management: An automatic network management system can detect migration of VMs and their new locations and adjust network configurations accordingly.

#2 Virtualized Networks: Network virtualization is similar to memory virtualization on computers. A centralized or distributed service control plane is deployed to abstract and encapsulate the physical network into logical networks. This service control plane shields complex protocol interactions on the physical network from upper-layer applications, and provides simple virtual network interfaces for applications.

#3 Centralized Management and Control: Traditional networks are static, and administrators only need to maintain the configurations of each network entity separately. Cloud networks are dynamic, and multiple network entities must be managed and scheduled together. Therefore, in the cloud computing era, a centralized management and control platform is required to manage the entire network as a whole, but not necessarily on a per-device basis.

Open Network

To protect initial investment and avoid restrictions of proprietary technologies, enterprises also need an open network with the following features:

#1 A service platform that provides open APIs to the application layer. For example, standard Openflow/SDN interfaces make network capabilities such as forwarding, QoS, and security control capabilities open to application platforms, such as OpenStack.

#2 A standard architecture that is open to third-party value-added services. For example, devices on the network have an open platform based on mainstream CPUs and OSs. This enables third-party firewall, IDS/IPS, application acceleration, and traffic analysis applications to run on this platform.

#3 Standard network protocols that ensure interconnection with devices from mainstream vendors.

The author is Head, Solutions, Huawei Enterprise India