Kathy Gibson at Huawei Connect in Shanghai — Organisations are looking to cloud computing as their digital transformation enabler, but there is still a tremendous amount of innovation taking place lower down the stack at the network level.
That’s the word from Ian Foo, director of product marketing at Huawei, who launched new networking technologies along with three built-for-purpose version of the Huawei CloudFabric.
The three version of CloudFabric, for hyperscale, high availability and high-performance computing (HPC), are the first in a what Foo expects to be a series of specific fabrics that will be launched to address specific use cases or industries.
Networking design is usually 80% commonality, with the balance catering for specific environments.
“So we are adding that 20% for specific environments,” he says. “For customers, this means most of the work is done, so it is easier to deploy and use these environments,” says Foo.
“You could do this stuff before, but now we are prepackaging it. So customers can order a generic CloudFabric,or they can request one that is configured for a specific network. Customers are telling us that they want their networks to be complete and quicker to deploy.”
CloudFabric for hyperscale environments delivers the ability to provide a highly scalable non-blocking network, meeting a range of scalability requirements.
This version of CloudFabric also adds SDN capabilities to improve deployments, and tiered hierarchical data centre capabilities.
Network administrators are able to join disparate pools of resources, so disparate data centres can seamlessly share resources.
This version of CloudFabric includes a major technological breakthrough, Foo adds. “It is underpinned by the launch of the industry’s first 100GE two-fibre interconnection. This means you need 75% fewer fibres to support the same capacity; or with the same number of fibre, you can quadruple the capacity in the data centre.
“This innovation allows is to provide interconnection from leaf to spine with a single pair of fibre.”
Other product enhancements include extensions to the switch that add end-to-end buffer capabilities. Controller capabilities have been advanced, and an SDN (software-defined networking) controller nnow implements service deployments within minutes.
“This allows us to reduce deployment costs and operating expenses (opex) by 35%.”
The high-availability version of CloudFabric offers zero service interruptions, along with service provisioning anytime and anywhere.
Other features of this solution include faster provisioning, internal security isolation, intelligent O&M, and increased reliability and availability.
The high availability fabric can reduce failure detection time by an order of magnitude, Foo adds. Administrators are able to compartmentalise the network down to the virtual machine for granular security.
High availability also translates into high visibility, so this version of CloudFabric offer fast fault isolation and prevention with FabricInsight. This collects data from all the components of the networks and models for abnormalities to prevent failure, giving adminstrators visibility down to the application within the network.
The HPC version of CloudFabric adds automation and HPC workloads and scalability. It is designed as a lossless network, with nanosecond latency.
A combination of hardware and software virtualisation helps congestion management to perform better to packages are queued for efficiently.
Because CloudFabric is software-defined, it is plastic and easy to use, Foo says. This can reduce job completion times by up to 40%, while the 100GE spine-leaf Ethernet increased network speed five-fold.
“SDN (software-defined networking) auto-provisioning means that HPC as a service (HPCaaS) is now possible,” says Foo.
Another technology innovation underpinning CloudFabric for HPC is what Foo describes as the industry’s highest-performance router, the NE9000-8, offering 4Tb per slot expandable to 8Tb, eight slots and 32Tb per chassis.
“The router complements the switching platform so we can scale between data centres, allowing seamless pooling of resources over fibre. The lower latency and higher throughput make it possible to treat them as one resource.”