It’s a concept that seems to make a lot of sense. The idea is that business customers will want the same flexibility from the network supporting cloud services that they get from the cloud itself.
Some people call this concept network-as-a-service, and it was a central idea of two presentations at the Cloud Services Summit – one from Sandeep Agrawal of Cisco and another from Milan Todorovic of N-Com, an engineering consulting firm.
Cisco’s Elastic Infrastructure
Agrawal described Cisco’s first-hand experiences over a four-year period with what the company calls “Cisco Elastic Infrastructure Services.” Cisco is famous for trying out new technologies to support its own internal operations and this is an example of that approach.
The company’s initial move was to replace its traditional data center environment based on a single application per server with a virtualized environment in which multiple applications could share a server. Using this approach, Cisco was able to reduce the time required to deliver new applications from as much as eight weeks to as little as two weeks. In addition, the company reduced its total cost of ownership 37%.
Cisco’s next move was to orchestrate capabilities on top of its virtualized infrastructure to enable new services to be provisioned on the fly, which generated an additional 27% reduction in the total cost of ownership, as well as reducing the speed of new service delivery to just 15 minutes.
More recently Cisco worked with a couple of telcos to orchestrate the provisioning of the data center in combination with the provisioning of the network. Using this approach, “the time and effort to create a service went down 80%,” Agrawal said.
If telcos can automate their network provisioning in combination with their cloud service delivery, “their value as a service provider will shine,” Agrawal said. He also noted that certain cloud services should be particularly well suited to this approach – including virtual desktop, collaboration/ conferencing, video delivery and hybrid public/private cloud services such as disaster recovery.
He noted, for example, that in a disaster recovery scenario, a business customer would need to move huge amounts of data from its private cloud data center to the telco’s public cloud infrastructure. Using the dynamic network provisioning he outlined, the bandwidth available to the customer would expand as needed to support that requirement.
Agrawal envisions a four-layer network infrastructure to support such capabilities. The four layers would include an application and service workflow, resource architecture software, virtualization infrastructure (which would be system-level software) and network gear.
Software defined networking
What Agrawal described sounds a lot like the approach behind software defined networking – a concept that is getting a lot of attention from large telcos – and in his presentation, Todorovic offered an excellent explanation of the concept behind software defined networking and how SDN could be used to enhance cloud services by automating network control.
As Todorovic explained, the idea behind software defined networking is to separate network control from individual network elements such as switches, routers and firewalls, instead moving that control to a new network device called a network controller.
“Today network elements are independent and make decisions based on what they know and limited information from their neighbors,” said Todorovic. “The elements don’t understand the state of the network as a whole.”
SDN aims to “decouple forwarding and decision-making functions” by taking decision-making out of individual network elements and moving it to the centralized controller, which has knowledge of the network as a whole, Todorovic said.
Todorovic referenced the OSI seven-layer networking model, noting that SDN initially aims to virtualize layers 4-6 so that new layer can talk to the layer 2-3 boxes that comprise today’s communication networks. Ultimately layers 2-3 also will be virtualized, essentially leaving only a single layer between the application at layer 7 and the physical media at layer 1, he said.
When this occurs, network elements will become generic and will essentially be processors with memory, Todorovic predicted. These network elements will be able to be programmed to take on the functions of a router, firewall or other device as needed, he said.
If, for example, multiple end users were requesting the same video stream from an over-the-top provider, the network controller could provision a “big pipe” to move content to a generic network element, which would be programmed to act as a router and a caching server.
Alternatively, if a business customer were to need to move its cloud data from one data center to another, the network controller could reconfigure a new network to support connectivity to the new data center.
It’s worth noting, though, that some of the ideas Agrawal and Todorovic described will take time to implement. In the meantime, Agrawal brought audience members back to present day realities. He essentially advised the audience to walk before running, cautioning anyone who has not yet offered some type of cloud service from jumping into network-as-a-service. Instead, he advised anyone just beginning to launch cloud services to consider infrastructure-as-a-service.
“There are great tools for that,” he said.