The first time I heard about Grid Computing I was still up at school in the late 90s which, not surprisingly, was around the same time I was engrossed in my parallel computing course. All the talk of IPC and modulo arithmetic joined in an unholy matrimony with massive amounts of Mountain Dew and 3am whiteboarding in the labs, the offspring of which was a simple understanding of why and when to use parallel computing.

Fast forward to today and Silicon Valley’s VCs have eliminated all but the most trendy of ideas to back, the recent darlings being so-called Cloud computing. So what is Cloud Computing? Supposedly The Cloud is an ultra-scalable architecture that has implicit redundancy so no one ever has to go through the painful process of upgrading hardware; is this not the panacea of Datacenter logistics – completely automated vertical scalability?

The Promise
Current providers of Cloud provisioning software, such as 3Tera, promise less cost, ease of maintenance, and simple scalability. Need a new server? Just click a few buttons and you have a new virtual server, possibly pre-configured with software and even your specific settings.

The Failure
While having a large and dynamic infrastructure is appealing, there are inherent problems with the current approaches, the most obvious of which is the lack of failure abstraction. While products like VMWare have the ability to provide fail-over automatically, the fact that offerings like Applogic and EC2 do not provide such capabilities out-of-the-box is very telling about the underlying architecture. While many can argue that Xen-based hypervisors can be instrumented to failover, the fact that the technology does not, as a matter of its DNA, provide failover is The Failure. In order to provide the promises marketing for the various grid services has suggested, not only do we need to scale easily, but we need reliability; you cannot provide one and not the other.

Commentary
We all know that technology changes rapidly, but vendors and pundits fail when they speak as if change is good just for the sake of itself. Recently we’ve seen many old architectures rehashed: mainframe dumb-terminals have become remote desktops and terminal services, talk of Ethernet’s limitations being replaced with a protocol that involves a “Token” and the new SMP craze disguised as multi-core processing. It’s time for some real progress. It may be in the same way virtualization has brought us the Cloud that the Cloud itself may be an intermediate step to something better. At the very least let’s hope that the next step on the ladder is up, not down.

So with so many moving parts to consider when getting services what should you look out for? Here are a few crucial things to keep in mind:

Global Routing
While many lay claim to how global the Internet is, those of us that operate backend systems know that there are plenty of caveats. The reachability of your server, and subsequently your services, is dependent upon how many peers your upstream provider has, and in turn, how many peers their peers have. As an example, in my area of the globe a company called Comcast provides residential broadband services. While I have one of our datacenters no more then 15 miles away, the best way to get a fast connection between my house and our local DC’s servers is to do some research and find out who Comcast peers with. It turns out that a nasty little secret is that, at least in my area, Comcast’s backbone is really Cogent, so we have a Cogent line coming into out local DC. By doing this the connection from my house to our local DC never leaves Cogent’s network, except for the edge equipment. Now there are always exceptions: if a link goes down for example the network gear will route around the failure, which may take my connection half-way across the country, but the primary route will typically be the shortest and least-expensive.

Business-Centric Infrastructure
Similar to many industries there are retail(consumer) and business(commercial) facets to the hosting industry, and as such providers tailor their services toward a particular audience. For businesses it’s important to find a provider that is aware of the requirements a business demands such as redundant components, high-performance network, reliable support contacts and value-added services. Today, more so then even 12 months ago, reliable power(main and backups) is high on the list of features to examine. It doesn’t hurt if a provider has current or previous experience in your industry either.

Managed Services
Perhaps the most import type of hosting for small and large businesses alike includes various forms of management, which indicates there are resources on-staff that are experts in a particular technology. There are a few handfuls of reliable hosting providers that can, for example, provision and host your server, manage the network and manage the software on your server. Most times the end-users of your server(i.e. your clients, your team, your internal users) are still your responsibility, but things like OS, network and backup configurations are handled by your service provider. The allure of managed services are two-fold: time savings for you and your team due to not worrying about the server and cost savings through having access to experts nearly immediately if issues ever arise.