If an organization makes the decision to launch a private cloud, what steps must be taken to pull off the plan? Cloud veterans and consultants cite five key moves.
“The way you go about designing a cloud network for a data center is quite different from when you are putting together a traditional campuswide network,” says Tony West, Cisco vice president and chief technology officer.
Begin by considering capacity. West advises that organizations estimate likely traffic volumes to and from the private cloud, based on the number of users and cloud servers that will be running. “Then make sure that you’ve got plenty of room for expansion, because one thing about the cloud is that any guesses about capacity are almost always wrong on the low end,” he explains.
Speed is the next factor to evaluate. “Increasingly, we are seeing compute and storage traffic pushing networks from 1 gigabit per second in the direction of 10 gigs,” West reports.
Running highly compatible hardware — all blade servers based on Intel Xeon processors, for example — allows IT managers to easily achieve the dynamic provision and resource-pooling goals of cloud computing.
The widespread server virtualization efforts of the past few years have helped many enterprises reduce physical hardware requirements, increase the utilization rates of the remaining devices and alleviate some power demands. Now organizations are looking beyond consolidation to the next big wave of virtualization, which spans servers, storage systems and networks.
“The idea now is to create portable workloads that I can put on any server at any time,” says Anil Desai, a consultant who specializes in cloud computing and virtualization. “I can move workloads while they are running, I can change the amount of memory they have and I can change which networks they are connected to. And I’m able to do all of this dynamically rather than shutting down and rebooting a server or physically adding memory to a machine.”
Cloud practitioners should look to specialized tools designed to keep workloads evenly balanced among all individual servers in the private cloud, while assuring that each application has the power it needs to meet prevailing demands. For example, a virtual machine may be designated to run on a high-performance server during regular working hours and then automatically move to a less powerful platform in the evening, when demand levels drop.
Clouds can accommodate many common types of enterprise services, such as e-mail systems, which do not require the full power of dedicated servers and related resources. More complex applications, such as large-scale enterprise resource planning (ERP) systems, are designed to run by themselves with their own resources.
How can IT managers make their final determinations? “Create an application map,” recommends Dhritiman Dasgupta, senior director of product marketing at Juniper Networks. “Take stock of all the applications you have today and the ones you expect to add in the next two to three years. Then, organize them according to which ones you will want to keep in-house, move to a public cloud or run partially within and outside the infrastructure.”