Hosting market segmentation – What’s the story? (I)

The hosting market offers us surprises along the way. Its structure is changing as the web changes, but as any market, it does have some consistent characteristics that we observed over the years. We’ve concentrated on an analysis meant to help those interested in understanding this market as well as those who are trying to figure out what’s all about with the Cloud, hosting, IaaS, PaaS, SaaS, Big Data as service providers.

The overall model is a so called Longtail model (Paretto distribution), as many other markets have. We split it into several categories: Shared, VDS+VPS, Dedicated servers. We overlaid Cloud offerings to find that the Cloud, in its various incarnations, covers many traditional segments.

We’ll also make references to a research made by Freeform Dynamics and sponsored by Microsoft, depicting another segmentation based on workload type and frequency.

At the end, we’ll try to point out different application types to different vendor types.

Hosting market segmentation based on “size”

We define the “size” of client as a fairly vague term derived from a combination of incoming traffic and required resources to handle a unit of traffic (say a unique visitor).





fig 1. Hosting market segment per size

There are a lot of “small” clients and a few “big” ones. The big ones usually require more complex solutions with caching, distributed databases and more intricate architectures, whereas the small ones can sometimes be handled by the hundreds on a single 1U server. 




Clients/ 1U

Shared up to 100k v/mth Shared is traditionally based on LAMP stack or ASP.NET. In recent years, with the advent of the cgroups technology in the Linux kernel, the clients are better isolated in “containers” but there is no virtualization involved usually. 300
VPS – Virtual private servers up to 300k v/mth VPSs are usually built with a shared kernel technology, mostly OpenVZ/Virtuozzo. It is not real virtualization as all the “VMs” run in the same Linux/Windows kernel even if the containers appear as individual machines to the client. It’s a more evolved form of cheroot. 50
VDS – Virtual dedicated Servers up to 500k v/mth VDS are built with real virtualization technology; the VMS are basically simulations of real hardware and the client should have the possibility to install in theory any operating system. This is done usually with VMware, Xen and KVM. 15
Dedicated Servers up to 2M v/mth Dedicated servers have been around for ages. They serve for one single client. The client can install a hypervisor of choice in order to further split the machine, but usually they run on bare metal. The provider usually pre-installs the OS for the customer. 1
Complex above 2M v/mth Complex solutions are usually a combination of services ranging from shared virtualized infrastructure to dedicated servers coupled with loadbalancers, firewalls, shared or dedicated storage etc. They involve highly skilled admins that build and maintain these infrastructures.

Table 1 – Summary of traditional hosting market segments based on size. Note that the “size” term is very relative and by no means accurate.


With the advent of the “Cloud”, things started to blur mostly because the Cloud is an umbrella term covering different aspects.

It is important to remember that the needs of the clients stayed more or less the same even though the industry now has to develop its product portfolio.

What has changed is the architecture of applications clients are starting to use. Failure to deliver a proper vertical scaling to the hosting industry, mostly due to the thermal barrier of CPU design, has given rise to new class of software architectures which scale horizontally and make use of the inherent parallelism of web applications. This, in turn, pushes the hosting industry to deliver services which fit this new architecture design.

We’ll cover the Cloud in more details later on as it introduces segmentation on its own.

Stay tuned for the next step of our hosting segmentation review – Market segmentation based on workload types.