Private Cloud 35
Building a Resilient Network for Your Cloud
Mel Beckman
October 3, 2013

We’ve been discussing the complexity of designing a network for your private cloud. Any virtual network for private cloud operations must be lossless, resilient, converged, edge-configurable, and software defined. Last time, we looked at the necessity of designing a lossless network. This week, we’ll look at the need for resiliency and convergence.

Resilient. The old term for this trait was fault-tolerant. But usually that meant tolerance for one fault, as in "Our fault-tolerant architecture has no single points of failure." That's not good enough for a private cloud. With many diverse tenants in your cloud, all expecting to stay nicely isolated from one another and online five nines of the time, your cloud network needs not just to tolerate faults (plural), but to circumvent them in ways that affect the fewest workloads.

When the network can't work around one or more faults, it must "degrade gracefully." A great many "fault-tolerant" networks survive the first failure but then take down the entire infrastructure when a second failure occurs. When shopping for resilience, ensure that failures of several components can't cascade. One key to graceful degradation is a management network that is isolated from production data. Technologies such as Microsoft server clustering can achieve these ends. Microsoft Hyper-V Network Virtualization is a resilience solution that lets your network gracefully tolerate multiple failures.

Converged. Legacy data centers sport two networks: one for application transactions between servers and users (i.e., the data network) and one over which applications access files on virtualized storage devices (i.e., the SAN). Each of these networks traditionally uses different technologies, optimized for one role or the other. But this single-taskedness comes at a very high cost in nonsharable components, incompatible interconnection technologies, and replicated management skills. A converged network meets objectives for both these legacy networks but uses a common technology (lossless Ethernet) that brings many economies of scale and a huge reduction in administrative labor. In a converged network, a port is a port: You designate how you want to use it at the time you plug a server or SAN device into it. The converged network has enough intelligence to separate the two kinds of traffic and ensure that each gets the appropriate priority and bandwidth. HP Virtual Connect technology perfectly illustrates the advantages of convergence, letting you slice network bandwidth capacity into arbitrarily sized channels that scale up to 40Mbps, and allocate those channels to any device, server, or virtual machine that you choose.

Next week, we’ll discuss why your network should be edge-configurable and software-defined.