Private Cloud 6
Designing a Network for Your Private Cloud
Mel Beckman
September 24, 2013

Choosing a server architecture is a natural first step for deploying your private cloud, as I discussed in the previous installment. After all, servers do the actual work in your cloud, so you need to select gear that meets your workload objectives. Everything else is secondary.

There are three more shopping stops before you're ready to power up your cloud, however. You need to buy networking, storage, and management components, in that order. Ultimately, your cloud network is the interconnecting mesh between all other cloud pieces. Storage and management components follow networking because ultimately you'll base storage and management decisions on the kind of network that you deploy.

You probably won't be amused to learn that shopping for a virtual network is the most difficult phase in building your cloud. Whereas servers simply virtualize CPU, memory, and I/O in a self-contained physical box using technology that's more than 40 years old, virtual networking is a relatively new technology. Virtual networking encompasses hundreds, even thousands, of interconnecting devices and cables. Your cloud's virtual network must by closely integrated with virtual hosts at the hypervisor level. And there are several hypervisors for network vendors to support, a constantly changing target that makes the job particularly challenging.

Despite all this complexity, there are several essential attributes that you want in any virtual network that is intended for cloud operations. A virtual network for private clouds needs to be (in no particular order) lossless, resilient, converged, edge-configurable, and software defined. Over the next few posts, I’ll take a look at each of these necessities.

Lossless. Most traditional data center networks lose packets. A lot of them. You tend not to notice because traffic is rarely at capacity for any one component, and because higher layer protocols, such as TCP/IP, resend lost packets. Therefore, applications are only delayed infrequently, and only slightly.

However, a private cloud, which aims to lower IT costs through higher efficiencies and economies of scale, can't tolerate loss. Early cloud pioneers found this out the hard way when small network problems and localized congestion cascaded into full-blown, multiday regional outages. As a result, network vendors have developed a set of technologies and protocols that make cloud networks truly lossless.

Vendors use a variety of approaches, both proprietary and open, but experience shows that open standards are usually a better choice over proprietary ones. Openness puts more brains on the problem and gives customers more choices. Vendors also realize that they can't count on technology locking to keep customers and so must truly innovate. Shop for products that support open specifications such as Data Center Bridging (DCB) and Transparent Interconnection of Lots of Links (TRILL). A good example of an open lossless network architecture is HP Intelligent Resilient Framework (IRF), which is based on standards such as TRILL and IS-IS inter-system routing.

Next time, we’ll look at the need for resiliency and a converged network.