Archive for September, 2011

Is IaaS fading away?

Wednesday, September 28th, 2011
Two weeks ago we had a really great panel at the Swedish Cloud Day, with participation from OpenNebula, Google, Berkeley and Ericsson (myself). We discussed quite some issues, but one kept bugging me ever since: are we seeing a decline of Infrastructure as a Service (IaaS), while PaaS and SaaS keep becoming dominant?
To answer that question, one has to look at when IaaS is really needed and by whom. One obvious use case is virtualization of existing applications, such as enterprise IT: moving existing enterprise services over to a cloud provider will require IaaS, for the simple reason that any other approach would mean a redesign of the existing software. In this case, IaaS is really a tool to support legacy applications, used by enterprise administrators rather than IT service consumers; it turns a two-tier system (enterprise IT systems – systems users) into a three-tier one: cloud provider – enterprise IT systems administrators – systems users. The real users of IaaS are enterprise IT systems admins; end users will still see SaaS as their primary model of interaction with enterprise IT.
A second use case is for casual utilization – both privately as well as for e.g. scientific or high performance computations. At the end, this is again really just a shift of legacy applications to the cloud: the only reason the user chooses IaaS is because the application is already given, (s)he only needs the scale-up service of the cloud. If (s)he would start from scratch, PaaS would be a better option.
Platform as a Service has some obvious benefits. It takes away the burden of managing instances, placement, networking, OS stack etc and enables the cloud user to focus on the core issue: developing an application that can quickly scale up and down based on needs. All the key services are given and guaranteed to scale and be available, something the IaaS model cannot provide. In many ways, the PaaS model gives back what infinite processor performance scaling took away: the notion of unlimited, reliable, on demand computing power.
I think Google really has a point here and in the future – and for new applications – we will see an increasing usage of PaaS offerings. The key issue though will be provider lock-in and tight coupling to platform APIs, something that no software provider can adopt lightheartedly. I believe we will see an emergence of portable libraries that still lock you into one API, but at least a portable one 😉