Yesterday Lori MacVittie posted another thoughtful article, Cloud Computing: Architectural Limbo, where she highlights percived problems with the NIST description. I usually agree with her cloud posts, but this is a rare case where I think she is wrong.

Consider, for a moment, the stark reality of a realm with no real network boundaries offered by AWS in “Building three-tier architectures with security groups”:

“Unlike with traditional on-premise physical deployments, AWS’s virtualization of compute, storage, and network elements requires that you think differently about how to build network segregation into your projects. There are no distinct physical networks, no VLANs, and no DMZs.”

The post goes on to describe the means in which a secure, traditional three-tiered application architecture can be deployed using AWS security groups. This architecture is a fine approximation of the traditional, data center deployed architecture based on the available abstractions offered by AWS

Note the use of the term “approximation”. That’s important, because it’s indicative of one of the core issues with cloud today: the inability to replicate architecture.

You might be thinking that’s okay as long as you can replicate it using available services.

Actually, it is okay because it does work. AWS does provide logical and physical barriers, and while they are presented in a way that only mimics traditional networks, they do so to ease understanding through familiar concepts. Being different does not make it less secure. And I’ve lost count of the number of organizations that have successfully deployed this (admittedly basic) architecture and are running it in production environments. It works so well that we even teach it in the CSA Certificate of Cloud Security Knowledge (CCSK) classes we run.

One of the great joys of running in an IaaS environment is its bare simplicity. You don’t have the crutches of a vast array of technology to rely on. Instead you have to think about your real needs, instead of adding huge amounts of complexity because that’s how we’re doing it in-house today. It’s a prime opportunity to start over and avoid repeating the sins of the past.

The problem is that in order to fully deploy in the cloud you have to deploy an architecture that will be different from the one you currently maintain in the data center. What that ultimately entails is a separate and environment-specific set of processes, as well, that could quickly become operationally expensive. This is especially true when compliance enters the picture, and even more so when the regulations in question are those that focus on process (think SOX) and not just technological implementation.

While it is true that in many cases, different network architectures and security requirements cause differences in cloud architectures, that doesn’t necessarily mean that the applications residing on those architectures will be fundamentally different. In my experience most operations teams have little or no knowledge of how the underlying network architectures are laid out. It’s simply irrelevant, so long as the necessary ports are open. And this is the model offered by cloud providers like AWS.

As for separate and environment-specific sets of processes, this is just a red herring. Network and especially security teams already have to do this, especially in larger organizations. You could just as well make this argument about every application deployment, regardless of locale. This is just part of life, and any good IT shop should be familiar it.